July 2010 - FFmpeg-soc - lists.ffmpeg.org

[soc]: r5873 - libavfilter/checkout.sh
by stefano 27 Jul '10

27 Jul '10

Author: stefano Date: Tue Jul 27 18:28:44 2010 New Revision: 5873 Log: Make checkout.sh copy to ffmpeg/libavfilter the list of files explicitely stated in the script, rather than all the files indiscriminately. More robust. Modified: libavfilter/checkout.sh Modified: libavfilter/checkout.sh ============================================================================== --- libavfilter/checkout.sh Tue Jul 27 18:17:15 2010 (r5872) +++ libavfilter/checkout.sh Tue Jul 27 18:28:44 2010 (r5873) @@ -12,4 +12,21 @@ echo "patching ffmpeg" for diff in $(ls $(pwd)/diffs/*.diff); do patch -d ffmpeg -p0 -i $diff; done echo "copying files to libavfilter" -find $(pwd) -maxdepth 1 -type f -not -name $(basename $0) -exec cp {} ffmpeg/libavfilter \; +cp \ + vf_clone.c \ + vf_concatenate.c \ + vf_drawbox.c \ + vf_drawtext.c \ + vf_fade.c \ + vf_fifo.c \ + vf_fps.c \ + vf_hflip.c \ + vf_negate.c \ + vf_overlay.c \ + vf_rotate.c \ + vf_setpts.c \ + vf_split.c \ + vf_transpose.c \ + vsrc_movie.c \ + ffmpeg/libavfilter +

1 0

[soc]: r5872 - in libavfilter: checkout.sh diffs/00_configure.diff diffs/01_build.diff diffs/03_doc_filters.diff
by stefano 27 Jul '10

27 Jul '10

Author: stefano Date: Tue Jul 27 18:17:15 2010 New Revision: 5872 Log: Update to FFmpeg r24545 / libswscale r31782. Modified: libavfilter/checkout.sh libavfilter/diffs/00_configure.diff libavfilter/diffs/01_build.diff libavfilter/diffs/03_doc_filters.diff Modified: libavfilter/checkout.sh ============================================================================== --- libavfilter/checkout.sh Tue Jul 27 18:11:35 2010 (r5871) +++ libavfilter/checkout.sh Tue Jul 27 18:17:15 2010 (r5872) @@ -1,11 +1,11 @@ #! /bin/sh echo "checking out pristine ffmpeg" -svn checkout svn://svn.ffmpeg.org/ffmpeg/trunk/ ffmpeg -r24294 +svn checkout svn://svn.ffmpeg.org/ffmpeg/trunk/ ffmpeg -r24545 echo "downloading the corresponding version of swscale" cd ffmpeg/libswscale -svn up -r31746 +svn up -r31782 cd ../.. echo "patching ffmpeg" Modified: libavfilter/diffs/00_configure.diff ============================================================================== --- libavfilter/diffs/00_configure.diff Tue Jul 27 18:11:35 2010 (r5871) +++ libavfilter/diffs/00_configure.diff Tue Jul 27 18:17:15 2010 (r5872) @@ -1,8 +1,8 @@ Index: configure =================================================================== ---- configure (revision 24294) +--- configure (revision 24545) +++ configure (working copy) -@@ -167,6 +167,7 @@ +@@ -168,6 +168,7 @@ and libraw1394 [no] --enable-libdirac enable Dirac support via libdirac [no] --enable-libfaac enable FAAC support via libfaac [no] @@ -10,7 +10,7 @@ Index: configure --enable-libgsm enable GSM support via libgsm [no] --enable-libmp3lame enable MP3 encoding via libmp3lame [no] --enable-libnut enable NUT (de)muxing via libnut, -@@ -869,6 +870,7 @@ +@@ -872,6 +873,7 @@ libdc1394 libdirac libfaac @@ -18,7 +18,7 @@ Index: configure libgsm libmp3lame libnut -@@ -1010,6 +1012,7 @@ +@@ -1014,6 +1016,7 @@ llrintf local_aligned_16 local_aligned_8 @@ -26,7 +26,7 @@ Index: configure log2 log2f loongson -@@ -1376,6 +1379,7 @@ +@@ -1381,6 +1384,7 @@ udp_protocol_deps="network" # filters @@ -34,7 +34,7 @@ Index: configure movie_filter_deps="avfilter_lavf" avfilter_lavf_deps="avformat" -@@ -2559,6 +2563,7 @@ +@@ -2579,6 +2583,7 @@ check_struct "sys/time.h sys/resource.h" "struct rusage" ru_maxrss check_func inet_aton $network_extralibs check_func isatty @@ -42,7 +42,7 @@ Index: configure check_func ${malloc_prefix}memalign && enable memalign check_func mkstemp check_func ${malloc_prefix}posix_memalign && enable posix_memalign -@@ -2641,6 +2646,7 @@ +@@ -2661,6 +2666,7 @@ require libdirac libdirac_decoder/dirac_parser.h dirac_decoder_init $(pkg-config --libs dirac) && require libdirac libdirac_encoder/dirac_encoder.h dirac_encoder_init $(pkg-config --libs dirac) enabled libfaac && require2 libfaac "stdint.h faac.h" faacEncGetVersion -lfaac Modified: libavfilter/diffs/01_build.diff ============================================================================== --- libavfilter/diffs/01_build.diff Tue Jul 27 18:11:35 2010 (r5871) +++ libavfilter/diffs/01_build.diff Tue Jul 27 18:17:15 2010 (r5872) @@ -1,13 +1,13 @@ Index: libavfilter/Makefile =================================================================== ---- libavfilter/Makefile (revision 24294) +--- libavfilter/Makefile (revision 24545) +++ libavfilter/Makefile (working copy) @@ -1,10 +1,11 @@ include $(SUBDIR)../config.mak NAME = avfilter --FFLIBS = avcodec avutil swscale -+FFLIBS = avcodec avutil +-FFLIBS = avcodec avcore avutil swscale ++FFLIBS = avcodec avcore avutil +FFLIBS-$(CONFIG_SCALE_FILTER) += swscale FFLIBS-$(CONFIG_AVFILTER_LAVF) += avformat @@ -51,7 +51,7 @@ Index: libavfilter/Makefile OBJS-$(CONFIG_NULLSINK_FILTER) += vsink_nullsink.o Index: libavfilter/allfilters.c =================================================================== ---- libavfilter/allfilters.c (revision 24294) +--- libavfilter/allfilters.c (revision 24545) +++ libavfilter/allfilters.c (working copy) @@ -36,19 +36,32 @@ Modified: libavfilter/diffs/03_doc_filters.diff ============================================================================== --- libavfilter/diffs/03_doc_filters.diff Tue Jul 27 18:11:35 2010 (r5871) +++ libavfilter/diffs/03_doc_filters.diff Tue Jul 27 18:17:15 2010 (r5872) @@ -1,6 +1,6 @@ Index: doc/filters.texi =================================================================== ---- doc/filters.texi (revision 24294) +--- doc/filters.texi (revision 24545) +++ doc/filters.texi (working copy) @@ -39,6 +39,114 @@ @@ -265,19 +265,7 @@ Index: doc/filters.texi @section unsharp Sharpen or blur the input video. -@@ -229,6 +450,11 @@ - - Below is a description of the currently available video sources. - -+@section buffer -+ -+This input filter is used by the client application to feed pictures to -+the filter chain. See ffmpeg.c for an usage example. -+ - @section color - - Provide an uniformly colored input. -@@ -268,6 +494,27 @@ +@@ -308,6 +529,27 @@ "color=red@@0.2:qcif:10 [color]; [in][color] overlay [out]" @end example

1 0

[soc]: r5871 - mms/mmst.c
by spyfeng 27 Jul '10

27 Jul '10

Author: spyfeng Date: Tue Jul 27 18:11:35 2010 New Revision: 5871 Log: use mms.h. Modified: mms/mmst.c Modified: mms/mmst.c ============================================================================== --- mms/mmst.c Tue Jul 27 17:30:35 2010 (r5870) +++ mms/mmst.c Tue Jul 27 18:11:35 2010 (r5871) @@ -28,7 +28,7 @@ * [2]http://msdn.microsoft.com/en-us/library/bb643323.aspx */ -#include "avformat.h" +#include "mms.h" #include "internal.h" #include "libavutil/intreadwrite.h" #include "libavcodec/bytestream.h" @@ -83,50 +83,6 @@ typedef enum { /*@}*/ } MMSSCPacketType; -typedef struct { - int id; -}MMSStream; - -typedef struct { - int outgoing_packet_seq; ///< Outgoing packet sequence number. - char path[256]; ///< Path of the resource being asked for. - char host[128]; ///< Host of the resources. - - URLContext *mms_hd; ///< TCP connection handle - MMSStream streams[MAX_STREAMS]; - - /** Buffer for outgoing packets. */ - /*@{*/ - uint8_t *write_out_ptr; ///< Pointer for writting the buffer. - uint8_t out_buffer[512]; ///< Buffer for outgoing packet. - /*@}*/ - - /** Buffer for incoming packets. */ - /*@{*/ - uint8_t in_buffer[8192]; ///< Buffer for incoming packets. - uint8_t *read_in_ptr; ///< Pointer for reading from incoming buffer. - int remaining_in_len; ///< Reading length from incoming buffer. - /*@}*/ - - int incoming_packet_seq; ///< Incoming packet sequence number. - int incoming_flags; ///< Incoming packet flags. - - int packet_id; ///< Identifier for packets in the current stream. - unsigned int header_packet_id; ///< default is 2. - - /** Internal handling of the ASF header */ - /*@{*/ - uint8_t *asf_header; ///< Stored ASF header. - int asf_header_size; ///< Size of stored ASF header. - int header_parsed; ///< The header has been received and parsed. - int asf_packet_len; - int asf_header_read_size; - /*@}*/ - - int stream_num; ///< stream numbers. - int is_playing; -} MMSContext; - /** Create MMST command packet header */ static void start_command_packet(MMSContext *mms, MMSCSPacketType packet_type) {

1 0

[soc]: r5870 - in mms: mms.h mmsh.c
by spyfeng 27 Jul '10

27 Jul '10

Author: spyfeng Date: Tue Jul 27 17:30:35 2010 New Revision: 5870 Log: move MMSHContext into mms.h for ready to extract common field with MMSContext. Modified: mms/mms.h mms/mmsh.c Modified: mms/mms.h ============================================================================== --- mms/mms.h Tue Jul 27 17:06:25 2010 (r5869) +++ mms/mms.h Tue Jul 27 17:30:35 2010 (r5870) @@ -66,4 +66,27 @@ typedef struct { int stream_num; ///< stream numbers. int is_playing; } MMSContext; + +typedef struct +{ + URLContext *mms_hd; + uint8_t out_buffer[8192]; ///< Buffer for outgoing packet. + uint8_t in_buffer[8192]; //TODO, maybe reused by out_buffer. + uint8_t *read_in_ptr; + MMSStream streams[MAX_STREAMS]; + + uint8_t *asf_header; + int asf_header_size; + int asf_header_read_size; + int asf_data_remaining_len; + int asf_packet_len; + + + char location[1024]; + int seekable; + int stream_num; + int request_seq; + int chunk_seq; + int is_header_parsed; +}MMSHContext; #endif Modified: mms/mmsh.c ============================================================================== --- mms/mmsh.c Tue Jul 27 17:06:25 2010 (r5869) +++ mms/mmsh.c Tue Jul 27 17:30:35 2010 (r5870) @@ -18,7 +18,7 @@ * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ -#include "avformat.h" +#include "mms.h" #include "internal.h" #include "libavutil/intreadwrite.h" #include <string.h> @@ -68,33 +68,6 @@ static const char* mmsh_live_request = "Pragma: stream-switch-entry=%s\r\n" "Connection: Close\r\n\r\n"; -typedef struct { - int id; -}MMSStream; - -typedef struct -{ - URLContext *mms_hd; - uint8_t out_buffer[8192]; ///< Buffer for outgoing packet. - uint8_t in_buffer[8192]; //TODO, maybe reused by out_buffer. - uint8_t *read_in_ptr; - MMSStream streams[MAX_STREAMS]; - - uint8_t *asf_header; - int asf_header_size; - int asf_header_read_size; - int asf_data_remaining_len; - int asf_packet_len; - - - char location[1024]; - int seekable; - int stream_num; - int request_seq; - int chunk_seq; - int is_header_parsed; -}MMSHContext; - static int mmsh_close(URLContext *h) { MMSHContext *mms = (MMSHContext *)h->priv_data;

1 0

[soc]: r5869 - mms/mmsh.c
by spyfeng 27 Jul '10

27 Jul '10

Author: spyfeng Date: Tue Jul 27 17:06:25 2010 New Revision: 5869 Log: use asf_header_parser() instead of the same function in mmsh.c. because that one is more maturity. Modified: mms/mmsh.c Modified: mms/mmsh.c ============================================================================== --- mms/mmsh.c Tue Jul 27 16:54:41 2010 (r5868) +++ mms/mmsh.c Tue Jul 27 17:06:25 2010 (r5869) @@ -102,54 +102,67 @@ static int mmsh_close(URLContext *h) url_close(mms->mms_hd); av_freep(&h->priv_data); av_freep(&mms->asf_header); - //TODO free other alloced mem. return 0; } -static int asf_header_parser(MMSHContext * mms) +static int asf_header_parser(MMSHContext *mms) { uint8_t *p = mms->asf_header; uint8_t *end; - int flags, stream_id, real_header_size; + int flags, stream_id; mms->stream_num = 0; if (mms->asf_header_size < sizeof(ff_asf_guid) * 2 + 22 || - memcmp(p, ff_asf_header, sizeof(ff_asf_guid))) - return -1; + memcmp(p, ff_asf_header, sizeof(ff_asf_guid))) { + av_log(NULL, AV_LOG_ERROR, + "Corrupt stream (invalid ASF header, size=%d)\n", + mms->asf_header_size); + return AVERROR_INVALIDDATA; + } - real_header_size = AV_RL64(p + sizeof(ff_asf_guid)); - end = mms->asf_header + real_header_size; + end = mms->asf_header + mms->asf_header_size; p += sizeof(ff_asf_guid) + 14; while(end - p >= sizeof(ff_asf_guid) + 8) { uint64_t chunksize = AV_RL64(p + sizeof(ff_asf_guid)); if (!chunksize || chunksize > end - p) { - dprintf(NULL, "chunksize is exceptional value:%"PRId64"!\n", chunksize); - return -1; + av_log(NULL, AV_LOG_ERROR, + "Corrupt stream (header chunksize %"PRId64" is invalid)\n", + chunksize); + return AVERROR_INVALIDDATA; } if (!memcmp(p, ff_asf_file_header, sizeof(ff_asf_guid))) { /* read packet size */ if (end - p > sizeof(ff_asf_guid) * 2 + 68) { mms->asf_packet_len = AV_RL32(p + sizeof(ff_asf_guid) * 2 + 64); if (mms->asf_packet_len <= 0 || mms->asf_packet_len > sizeof(mms->in_buffer)) { - dprintf(NULL,"Too large packet len:%d" - " may overwrite in_buffer when padding", mms->asf_packet_len); - return -1; + av_log(NULL, AV_LOG_ERROR, + "Corrupt stream (too large pkt_len %d)\n", + mms->asf_packet_len); + return AVERROR_INVALIDDATA; } } } else if (!memcmp(p, ff_asf_stream_header, sizeof(ff_asf_guid))) { flags = AV_RL16(p + sizeof(ff_asf_guid)*3 + 24); stream_id = flags & 0x7F; - if (mms->stream_num < MAX_STREAMS ) { + //The second condition is for checking CS_PKT_STREAM_ID_REQUEST packet size, + //we can calcuate the packet size by stream_num. + //Please see function send_stream_selection_request(). + if (mms->stream_num < MAX_STREAMS && + 46 + mms->stream_num * 6 < sizeof(mms->out_buffer)) { mms->streams[mms->stream_num].id = stream_id; mms->stream_num++; } else { - dprintf(NULL, "Too many streams.\n"); - return -1; + av_log(NULL, AV_LOG_ERROR, + "Corrupt stream (too many A/V streams)\n"); + return AVERROR_INVALIDDATA; } + } else if (!memcmp(p, ff_asf_head1_guid, sizeof(ff_asf_guid))) { + chunksize = 46; // see mmst.c references [2] section 3.4. This should be set 46. } p += chunksize; } + return 0; }

1 0

[soc]: r5868 - mms/mms.h
by spyfeng 27 Jul '10

27 Jul '10

Author: spyfeng Date: Tue Jul 27 16:54:41 2010 New Revision: 5868 Log: add mms.h for extracting the common code from mmst.c and mmsh.c, mms.c will be added later. Added: mms/mms.h Added: mms/mms.h ============================================================================== --- /dev/null 00:00:00 1970 (empty, because file is newly added) +++ mms/mms.h Tue Jul 27 16:54:41 2010 (r5868) @@ -0,0 +1,69 @@ +/* + * MMS protocol over HTTP + * Copyright (c) 2010 Zhentan Feng <spyfeng at gmail dot com> + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ +#ifndef AVFORMAT_MMS_H +#define AVFORMAT_MMS_H + +#include "avformat.h" + +typedef struct { + int id; +}MMSStream; + +typedef struct { + int outgoing_packet_seq; ///< Outgoing packet sequence number. + char path[256]; ///< Path of the resource being asked for. + char host[128]; ///< Host of the resources. + + URLContext *mms_hd; ///< TCP connection handle + MMSStream streams[MAX_STREAMS]; + + /** Buffer for outgoing packets. */ + /*@{*/ + uint8_t *write_out_ptr; ///< Pointer for writting the buffer. + uint8_t out_buffer[512]; ///< Buffer for outgoing packet. + /*@}*/ + + /** Buffer for incoming packets. */ + /*@{*/ + uint8_t in_buffer[8192]; ///< Buffer for incoming packets. + uint8_t *read_in_ptr; ///< Pointer for reading from incoming buffer. + int remaining_in_len; ///< Reading length from incoming buffer. + /*@}*/ + + int incoming_packet_seq; ///< Incoming packet sequence number. + int incoming_flags; ///< Incoming packet flags. + + int packet_id; ///< Identifier for packets in the current stream. + unsigned int header_packet_id; ///< default is 2. + + /** Internal handling of the ASF header */ + /*@{*/ + uint8_t *asf_header; ///< Stored ASF header. + int asf_header_size; ///< Size of stored ASF header. + int header_parsed; ///< The header has been received and parsed. + int asf_packet_len; + int asf_header_read_size; + /*@}*/ + + int stream_num; ///< stream numbers. + int is_playing; +} MMSContext; +#endif

1 0

[soc]: r5867 - mms/mmst.c
by spyfeng 27 Jul '10

27 Jul '10

Author: spyfeng Date: Tue Jul 27 16:46:26 2010 New Revision: 5867 Log: synchronize with ffmpeg/trunk because Ronald has modified it. Modified: mms/mmst.c Modified: mms/mmst.c ============================================================================== --- mms/mmst.c Tue Jul 20 17:30:36 2010 (r5866) +++ mms/mmst.c Tue Jul 27 16:46:26 2010 (r5867) @@ -21,10 +21,10 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ -/*References +/* References * MMS protocol specification: * [1]http://msdn.microsoft.com/en-us/library/cc234711(PROT.10).aspx - *asf specification. Revision 01.20.03. + * ASF specification. Revision 01.20.03. * [2]http://msdn.microsoft.com/en-us/library/bb643323.aspx */ @@ -170,8 +170,11 @@ static int send_command_packet(MMSContex // write it out. write_result= url_write(mms->mms_hd, mms->out_buffer, exact_length); if(write_result != exact_length) { - dprintf(NULL, "url_write returned: %d != %d\n", - write_result, exact_length); + av_log(NULL, AV_LOG_ERROR, + "Failed to write data of length %d: %d (%s)\n", + exact_length, write_result, + write_result < 0 ? strerror(write_result) : + "The server closed the connection"); return AVERROR_IO; } @@ -264,117 +267,132 @@ static MMSSCPacketType get_tcp_server_re MMSSCPacketType packet_type= -1; for(;;) { - if((read_result= url_read_complete(mms->mms_hd, mms->in_buffer, 8))==8) { - // handle command packet. - if(AV_RL32(mms->in_buffer + 4)==0xb00bface) { - mms->incoming_flags= mms->in_buffer[3]; - read_result= url_read_complete(mms->mms_hd, mms->in_buffer+8, 4); - if(read_result == 4) { - int length_remaining= AV_RL32(mms->in_buffer+8) + 4; - int hr; - - dprintf(NULL, "Length remaining is %d\n", length_remaining); - // read the rest of the packet. - if (length_remaining < 0 - || length_remaining > sizeof(mms->in_buffer) - 12) { - dprintf(NULL, "Incoming message len %d exceeds buffer len %d\n", - length_remaining, sizeof(mms->in_buffer) - 12); - return -1; - } - read_result = url_read_complete(mms->mms_hd, mms->in_buffer + 12, - length_remaining) ; - if (read_result == length_remaining) { - packet_type= AV_RL16(mms->in_buffer+36); - } else { - dprintf(NULL, "read for packet type failed%d!\n", read_result); - return -1; - } - hr = AV_RL32(mms->in_buffer + 40); - if (hr) { - dprintf(NULL, "The server side send back error code:0x%x\n", hr); - return -1; - } - } else { - dprintf(NULL, "read for length remaining failed%d!\n", read_result); - return -1; - } + read_result = url_read_complete(mms->mms_hd, mms->in_buffer, 8); + if (read_result != 8) { + if(read_result < 0) { + av_log(NULL, AV_LOG_ERROR, + "Error reading packet header: %d (%s)\n", + read_result, strerror(read_result)); + packet_type = SC_PKT_CANCEL; } else { - int length_remaining; - int packet_id_type; - int tmp; + av_log(NULL, AV_LOG_ERROR, + "The server closed the connection\n"); + packet_type = SC_PKT_NO_DATA; + } + return packet_type; + } - // note we cache the first 8 bytes, - // then fill up the buffer with the others - tmp = AV_RL16(mms->in_buffer + 6); - length_remaining = (tmp - 8) & 0xffff; - mms->incoming_packet_seq = AV_RL32(mms->in_buffer); - packet_id_type = mms->in_buffer[4]; - mms->incoming_flags = mms->in_buffer[5]; + // handle command packet. + if(AV_RL32(mms->in_buffer + 4)==0xb00bface) { + int length_remaining, hr; - if (length_remaining < 0 - || length_remaining > sizeof(mms->in_buffer) - 8) { - dprintf(NULL, "Incoming data len %d exceeds buffer len %d\n", - length_remaining, sizeof(mms->in_buffer)); - return -1; - } - mms->remaining_in_len = length_remaining; - mms->read_in_ptr = mms->in_buffer; - read_result= url_read_complete(mms->mms_hd, mms->in_buffer, length_remaining); - if(read_result != length_remaining) { - dprintf(NULL, "read_bytes result: %d asking for %d\n", - read_result, length_remaining); - return -1; - } else { - // if we successfully read everything. - if(packet_id_type == mms->header_packet_id) { - packet_type = SC_PKT_ASF_HEADER; - // Store the asf header - if(!mms->header_parsed) { - void *p = av_realloc(mms->asf_header, - mms->asf_header_size - + mms->remaining_in_len); - if (!p) { - av_freep(&mms->asf_header); - return AVERROR(ENOMEM); - } - mms->asf_header = p; - memcpy(mms->asf_header + mms->asf_header_size, - mms->read_in_ptr, - mms->remaining_in_len); - mms->asf_header_size += mms->remaining_in_len; - } - // 0x04 means asf header is sent in multiple packets. - if (mms->incoming_flags == 0x04) - continue; - } else if(packet_id_type == mms->packet_id) { - packet_type = SC_PKT_ASF_MEDIA; - } else { - dprintf(NULL, "packet id type %d is old.", packet_id_type); - continue; - } - } + mms->incoming_flags= mms->in_buffer[3]; + read_result= url_read_complete(mms->mms_hd, mms->in_buffer+8, 4); + if(read_result != 4) { + av_log(NULL, AV_LOG_ERROR, + "Reading command packet length failed: %d (%s)\n", + read_result, + read_result < 0 ? strerror(read_result) : + "The server closed the connection"); + return read_result < 0 ? read_result : AVERROR_IO; } - // preprocess some packet type - if(packet_type == SC_PKT_KEEPALIVE) { - send_keepalive_packet(mms); - continue; - } else if(packet_type == SC_PKT_STREAM_CHANGING) { - handle_packet_stream_changing_type(mms); - } else if(packet_type == SC_PKT_ASF_MEDIA) { - pad_media_packet(mms); + length_remaining= AV_RL32(mms->in_buffer+8) + 4; + dprintf(NULL, "Length remaining is %d\n", length_remaining); + // read the rest of the packet. + if (length_remaining < 0 + || length_remaining > sizeof(mms->in_buffer) - 12) { + av_log(NULL, AV_LOG_ERROR, + "Incoming packet length %d exceeds bufsize %zu\n", + length_remaining, sizeof(mms->in_buffer) - 12); + return AVERROR_INVALIDDATA; + } + read_result = url_read_complete(mms->mms_hd, mms->in_buffer + 12, + length_remaining) ; + if (read_result != length_remaining) { + av_log(NULL, AV_LOG_ERROR, + "Reading pkt data (length=%d) failed: %d (%s)\n", + length_remaining, read_result, + read_result < 0 ? strerror(read_result) : + "The server closed the connection"); + return read_result < 0 ? read_result : AVERROR_IO; + } + packet_type= AV_RL16(mms->in_buffer+36); + hr = AV_RL32(mms->in_buffer + 40); + if (hr) { + av_log(NULL, AV_LOG_ERROR, + "Server sent an error status code: 0x%08x\n", hr); + return AVERROR_UNKNOWN; } - return packet_type; } else { - if(read_result<0) { - dprintf(NULL, "Read error (or cancelled) returned %d!\n", read_result); - packet_type = SC_PKT_CANCEL; + int length_remaining; + int packet_id_type; + int tmp; + + // note we cache the first 8 bytes, + // then fill up the buffer with the others + tmp = AV_RL16(mms->in_buffer + 6); + length_remaining = (tmp - 8) & 0xffff; + mms->incoming_packet_seq = AV_RL32(mms->in_buffer); + packet_id_type = mms->in_buffer[4]; + mms->incoming_flags = mms->in_buffer[5]; + + if (length_remaining < 0 + || length_remaining > sizeof(mms->in_buffer) - 8) { + av_log(NULL, AV_LOG_ERROR, + "Data length %d is invalid or too large (max=%zu)\n", + length_remaining, sizeof(mms->in_buffer)); + return AVERROR_INVALIDDATA; + } + mms->remaining_in_len = length_remaining; + mms->read_in_ptr = mms->in_buffer; + read_result= url_read_complete(mms->mms_hd, mms->in_buffer, length_remaining); + if(read_result != length_remaining) { + av_log(NULL, AV_LOG_ERROR, + "Failed to read packet data of size %d: %d (%s)\n", + length_remaining, read_result, + read_result < 0 ? strerror(read_result) : + "The server closed the connection"); + return read_result < 0 ? read_result : AVERROR_IO; + } + + // if we successfully read everything. + if(packet_id_type == mms->header_packet_id) { + packet_type = SC_PKT_ASF_HEADER; + // Store the asf header + if(!mms->header_parsed) { + void *p = av_realloc(mms->asf_header, + mms->asf_header_size + mms->remaining_in_len); + if (!p) { + av_freep(&mms->asf_header); + return AVERROR(ENOMEM); + } + mms->asf_header = p; + memcpy(mms->asf_header + mms->asf_header_size, + mms->read_in_ptr, mms->remaining_in_len); + mms->asf_header_size += mms->remaining_in_len; + } + // 0x04 means asf header is sent in multiple packets. + if (mms->incoming_flags == 0x04) + continue; + } else if(packet_id_type == mms->packet_id) { + packet_type = SC_PKT_ASF_MEDIA; } else { - dprintf(NULL, "Read result of zero?!\n"); - packet_type = SC_PKT_NO_DATA; + dprintf(NULL, "packet id type %d is old.", packet_id_type); + continue; } - return packet_type; } + + // preprocess some packet type + if(packet_type == SC_PKT_KEEPALIVE) { + send_keepalive_packet(mms); + continue; + } else if(packet_type == SC_PKT_STREAM_CHANGING) { + handle_packet_stream_changing_type(mms); + } else if(packet_type == SC_PKT_ASF_MEDIA) { + pad_media_packet(mms); + } + return packet_type; } } @@ -392,8 +410,10 @@ static int mms_safe_send_recv(MMSContext } if ((type = get_tcp_server_response(mms)) != expect_type) { - dprintf(NULL,"Unexpected packet type %d with type %d\n", type, expect_type); - return -1; + av_log(NULL, AV_LOG_ERROR, + "Corrupt stream (unexpected packet type 0x%x, expected 0x%x)\n", + type, expect_type); + return AVERROR_INVALIDDATA; } else { return 0; } @@ -446,8 +466,12 @@ static int asf_header_parser(MMSContext mms->stream_num = 0; if (mms->asf_header_size < sizeof(ff_asf_guid) * 2 + 22 || - memcmp(p, ff_asf_header, sizeof(ff_asf_guid))) - return -1; + memcmp(p, ff_asf_header, sizeof(ff_asf_guid))) { + av_log(NULL, AV_LOG_ERROR, + "Corrupt stream (invalid ASF header, size=%d)\n", + mms->asf_header_size); + return AVERROR_INVALIDDATA; + } end = mms->asf_header + mms->asf_header_size; @@ -455,17 +479,20 @@ static int asf_header_parser(MMSContext while(end - p >= sizeof(ff_asf_guid) + 8) { uint64_t chunksize = AV_RL64(p + sizeof(ff_asf_guid)); if (!chunksize || chunksize > end - p) { - dprintf(NULL, "chunksize is exceptional value:%"PRId64"!\n", chunksize); - return -1; + av_log(NULL, AV_LOG_ERROR, + "Corrupt stream (header chunksize %"PRId64" is invalid)\n", + chunksize); + return AVERROR_INVALIDDATA; } if (!memcmp(p, ff_asf_file_header, sizeof(ff_asf_guid))) { /* read packet size */ if (end - p > sizeof(ff_asf_guid) * 2 + 68) { mms->asf_packet_len = AV_RL32(p + sizeof(ff_asf_guid) * 2 + 64); if (mms->asf_packet_len <= 0 || mms->asf_packet_len > sizeof(mms->in_buffer)) { - dprintf(NULL,"Too large packet len:%d" - " may overwrite in_buffer when padding", mms->asf_packet_len); - return -1; + av_log(NULL, AV_LOG_ERROR, + "Corrupt stream (too large pkt_len %d)\n", + mms->asf_packet_len); + return AVERROR_INVALIDDATA; } } } else if (!memcmp(p, ff_asf_stream_header, sizeof(ff_asf_guid))) { @@ -479,8 +506,9 @@ static int asf_header_parser(MMSContext mms->streams[mms->stream_num].id = stream_id; mms->stream_num++; } else { - dprintf(NULL, "Too many streams.\n"); - return -1; + av_log(NULL, AV_LOG_ERROR, + "Corrupt stream (too many A/V streams)\n"); + return AVERROR_INVALIDDATA; } } else if (!memcmp(p, ff_asf_head1_guid, sizeof(ff_asf_guid))) { chunksize = 46; // see references [2] section 3.4. This should be set 46. @@ -546,9 +574,9 @@ static int read_mms_packet(MMSContext *m int err = mms_safe_send_recv(mms, NULL, SC_PKT_ASF_MEDIA); if (err == 0) { if(mms->remaining_in_len>mms->asf_packet_len) { - dprintf(NULL, "Incoming packet" - "larger than the asf packet size stated (%d>%d)\n", - mms->remaining_in_len, mms->asf_packet_len); + av_log(NULL, AV_LOG_ERROR, + "Incoming pktlen %d is larger than ASF pktsize %d\n", + mms->remaining_in_len, mms->asf_packet_len); result= AVERROR_IO; } else { // copy the data to the packet buffer.

1 0

[PATCH] fix segfault in xiph depacketizer
by Josh Allmann 27 Jul '10

27 Jul '10

This fixes a segfault that sometimes occurs with live/broadcast streams, since the first packet we receive will not always be the start of a frame, in the case of fragmented frames. This leads to writing a null buffer. Josh

2 1

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 26 Jul '10

26 Jul '10

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "AMR-WB decoder". The branch, master has been updated via 4e9169a62e5fdb901d0f1a0a4a85613b575e7b2f (commit) via eff9bcbdb922197ef4720daa98bc6e01b8e4fb80 (commit) via fe78e3f003f346c235c38403442d8f4f9f541afd (commit) from 0ca8fed46c9ff9b50f5d51f2d1fa79d8583aedab (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit 4e9169a62e5fdb901d0f1a0a4a85613b575e7b2f Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Mon Jul 26 22:46:33 2010 -0300 Implement output upsampling to 16kHz, this finishes the lower band generation part, but not functional yet diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index d8a1068..46a97ff 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -26,10 +26,14 @@ #define LP_ORDER 16 ///< linear predictive coding filter order #define LP_ORDER_16k 20 ///< lpc filter order at 16kHz +#define UPS_FIR_SIZE 12 ///< upsampling filter size +#define SAMPLE_MEM 24 ///< number of stored past samples given by + ///< max(LP_ORDER, UPS_FIR_SIZE * 2) #define MIN_ISF_SPACING (128 / 32768.0) ///< minimum isf gap #define PRED_FACTOR (1.0 / 3.0) #define MIN_ENERGY -14.0 ///< initial innnovation energy (dB) #define ENERGY_MEAN 30.0 ///< mean innovation energy (dB) in all modes +#define PREEMPH_FAC 0.68 ///< factor used to de-emphasize synthesis #define AMRWB_SUBFRAME_SIZE 64 ///< samples per subframe at 12.8 kHz #define AMRWB_SFR_SIZE_OUT 80 ///< samples per subframe at 16 kHz @@ -38,8 +42,6 @@ #define AMRWB_P_DELAY_MAX 231 ///< maximum pitch delay value #define AMRWB_P_DELAY_MIN 34 -#define PREEMPH_FAC 0.68 ///< factor used to de-emphasize synthesis - /* Mode ordering is sensitive, do not change */ enum Mode { MODE_6k60 = 0, ///< 6.60 kbit/s @@ -1757,7 +1759,6 @@ static const float energy_pred_fac[4] = { 0.2, 0.3, 0.4, 0.5 }; /** impulse response filter tables converted to float from Q15 * used for anti-sparseness processing */ -// XXX: Not sure whether it is Q15 indeed static const float ir_filter_str[64] = { 0.615906, 0.295807, 0.099792, -0.104889, 0.087402, -0.159912, 0.048492, -0.041412, 0.018311, 0.118805, -0.045685, -0.021301, @@ -1802,6 +1803,35 @@ static const float hpf_400_coef[2][3] = { // 400 kHz cutoff filter { 1.787109375, -0.864257812, 0 } }; +/* Interpolation coefficients for 5/4 signal upsampling + * Table from the reference source was reordered for efficiency */ +static const float upsample_fir[4][24] = { + { -6.103516e-05, 7.324219e-04, -2.014160e-03, 4.150391e-03, + -7.263184e-03, 1.165771e-02, -1.776123e-02, 2.624512e-02, + -3.869629e-02, 5.877686e-02, -9.863281e-02, 2.314453e-01, + 9.348755e-01, -1.523438e-01, 7.861328e-02, -4.937744e-02, + 3.308105e-02, -2.252197e-02, 1.507568e-02, -9.765625e-03, + 5.859375e-03, -3.173828e-03, 1.403809e-03, -3.662109e-04 }, + { -2.441406e-04, 1.464844e-03, -3.784180e-03, 7.568359e-03, + -1.300049e-02, 2.062988e-02, -3.112793e-02, 4.589844e-02, + -6.781006e-02, 1.042480e-01, -1.815186e-01, 5.016479e-01, + 7.548828e-01, -2.094727e-01, 1.148071e-01, -7.348633e-02, + 4.956055e-02, -3.369141e-02, 2.246094e-02, -1.434326e-02, + 8.483887e-03, -4.455566e-03, 1.831055e-03, -4.272461e-04 }, + { -4.272461e-04, 1.831055e-03, -4.455566e-03, 8.483887e-03, + -1.434326e-02, 2.246094e-02, -3.369141e-02, 4.956055e-02, + -7.348633e-02, 1.148071e-01, -2.094727e-01, 7.548828e-01, + 5.016479e-01, -1.815186e-01, 1.042480e-01, -6.781006e-02, + 4.589844e-02, -3.112793e-02, 2.062988e-02, -1.300049e-02, + 7.568359e-03, -3.784180e-03, 1.464844e-03, -2.441406e-04 }, + { -3.662109e-04, 1.403809e-03, -3.173828e-03, 5.859375e-03, + -9.765625e-03, 1.507568e-02, -2.252197e-02, 3.308105e-02, + -4.937744e-02, 7.861328e-02, -1.523438e-01, 9.348755e-01, + 2.314453e-01, -9.863281e-02, 5.877686e-02, -3.869629e-02, + 2.624512e-02, -1.776123e-02, 1.165771e-02, -7.263184e-03, + 4.150391e-03, -2.014160e-03, 7.324219e-04, -6.103516e-05 } +}; + /* High band quantized gains for 23k85 in Q14 */ static const uint16_t qua_hb_gain[16] = { 3624, 4673, 5597, 6479, 7425, 8378, 9324, 10264, diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 2a62a17..d7deb39 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -72,7 +72,7 @@ typedef struct { uint8_t prev_ir_filter_nr; ///< previous impulse response filter "impNr": 0 - strong, 1 - medium, 2 - none float prev_tr_gain; ///< previous initial gain used by noise enhancer for thresold - float samples_in[LP_ORDER + AMRWB_SUBFRAME_SIZE]; ///< floating point samples + float samples_in[SAMPLE_MEM + AMRWB_SUBFRAME_SIZE]; ///< lower band floating point samples at 12.8kHz float demph_mem[1]; ///< previous value in the de-emphasis filter float hpf_31_mem[4], hpf_400_mem[4]; ///< previous values in the high-pass filters @@ -961,6 +961,33 @@ static void high_pass_filter(float *out, const float hpf_coef[2][3], } /** + * Upsample a signal by 5/4 ratio (from 12.8kHz to 16kHz) using + * a FIR interpolation filter. Uses past data from before *in address + * + * @param out [out] buffer for interpolated signal + * @param in [in] current signal data (length 0.8*o_size) + * @param o_size [in] output signal length + */ +static void upsample_5_4(float *out, const float *in, int o_size) +{ + const float *in0 = in - UPS_FIR_SIZE + 1; + int i; + + for (i = 0; i < o_size; i++) { + int int_part = (i << 2) / 5; + int frac_part = (i << 2) - 5 * int_part; + + if (!frac_part) { + out[i] = in[i]; + } else + out[i] = ff_dot_productf(in0 + int_part, upsample_fir[4 - frac_part], + UPS_FIR_SIZE << 1); + + out[i] *= 2.0; // upscale output + } +} + +/** * Calculate the high band gain based on encoded index (23k85 mode) or * on the lower band speech signal and the Voice Activity Detection flag * @@ -1068,7 +1095,7 @@ static void update_sub_state(AMRWBContext *ctx) memmove(&ctx->fixed_gain[0], &ctx->fixed_gain[1], 4 * sizeof(float)); memmove(&ctx->samples_in[0], &ctx->samples_in[AMRWB_SUBFRAME_SIZE], - LP_ORDER * sizeof(float)); + SAMPLE_MEM * sizeof(float)); } static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, @@ -1078,6 +1105,7 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, AMRWBFrame *cf = &ctx->frame; const uint8_t *buf = avpkt->data; int buf_size = avpkt->size; + float *buf_out = data; AMRFixed fixed_sparse = {0}; // fixed vector up to anti-sparseness processing float spare_vector[AMRWB_SUBFRAME_SIZE]; // extra stack space to hold result from anti-sparseness processing float fixed_gain_factor; // fixed gain correction factor (gamma) @@ -1172,22 +1200,22 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, pitch_enhancer(synth_fixed_vector, voice_fac); synthesis(ctx, ctx->lp_coef[sub], synth_exc, synth_fixed_gain, - synth_fixed_vector, &ctx->samples_in[LP_ORDER]); + synth_fixed_vector, &ctx->samples_in[SAMPLE_MEM]); /* Synthesis speech post-processing */ - de_emphasis(&ctx->samples_in[LP_ORDER], PREEMPH_FAC, ctx->demph_mem); - - high_pass_filter(&ctx->samples_in[LP_ORDER], hpf_31_coef, - ctx->hpf_31_mem, &ctx->samples_in[LP_ORDER]); + de_emphasis(&ctx->samples_in[SAMPLE_MEM], PREEMPH_FAC, ctx->demph_mem); - // XXX: the 5/4 upsampling for the lower band goes in here + high_pass_filter(&ctx->samples_in[SAMPLE_MEM], hpf_31_coef, + ctx->hpf_31_mem, &ctx->samples_in[SAMPLE_MEM]); + upsample_5_4(buf_out + sub * AMRWB_SFR_SIZE_OUT, + &ctx->samples_in[UPS_FIR_SIZE], AMRWB_SFR_SIZE_OUT); /* High frequency band generation */ - high_pass_filter(&ctx->samples_in[LP_ORDER], hpf_400_coef, - ctx->hpf_400_mem, &ctx->samples_in[LP_ORDER]); + high_pass_filter(&ctx->samples_in[SAMPLE_MEM], hpf_400_coef, + ctx->hpf_400_mem, &ctx->samples_in[SAMPLE_MEM]); - hb_gain = find_hb_gain(ctx, &ctx->samples_in[LP_ORDER], + hb_gain = find_hb_gain(ctx, &ctx->samples_in[SAMPLE_MEM], cur_subframe->hb_gain, cf->vad); scaled_hb_excitation(ctx, hb_exc, synth_exc, hb_gain); commit eff9bcbdb922197ef4720daa98bc6e01b8e4fb80 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Mon Jul 26 18:03:34 2010 -0300 Remove overflow treatment from synthesis (it was NB specific) diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 8768863..2a62a17 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -872,7 +872,6 @@ static void pitch_enhancer(float *fixed_vector, float voice_fac) /** * Conduct 16th order linear predictive coding synthesis from excitation - * Return a overflow detection flag * * @param ctx [in] pointer to the AMRWBContext * @param lpc [in] pointer to the LPC coefficients @@ -880,25 +879,17 @@ static void pitch_enhancer(float *fixed_vector, float voice_fac) * @param fixed_gain [in] fixed codebook gain for synthesis * @param fixed_vector [in] algebraic codebook vector * @param samples [out] pointer to the output speech samples - * @param overflow [in] 16-bit predicted overflow flag */ -static uint8_t synthesis(AMRWBContext *ctx, float *lpc, float *excitation, +static void synthesis(AMRWBContext *ctx, float *lpc, float *excitation, float fixed_gain, const float *fixed_vector, - float *samples, uint8_t overflow) + float *samples) { - int i; - - // if an overflow has been detected, the pitch vector is scaled down by a - // factor of 4 - if (overflow) - for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) - ctx->pitch_vector[i] *= 0.25; - ff_weighted_vector_sumf(excitation, ctx->pitch_vector, fixed_vector, ctx->pitch_gain[4], fixed_gain, AMRWB_SUBFRAME_SIZE); // emphasize pitch vector contribution in low bitrate modes - if (ctx->pitch_gain[4] > 0.5 && !overflow && ctx->fr_cur_mode <= MODE_8k85) { + if (ctx->pitch_gain[4] > 0.5 && ctx->fr_cur_mode <= MODE_8k85) { + int i; float energy = ff_dot_productf(excitation, excitation, AMRWB_SUBFRAME_SIZE); @@ -915,14 +906,6 @@ static uint8_t synthesis(AMRWBContext *ctx, float *lpc, float *excitation, ff_celp_lp_synthesis_filterf(samples, lpc, excitation, AMRWB_SUBFRAME_SIZE, LP_ORDER); - - // detect overflow - for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) - if (fabsf(samples[i]) > AMRWB_SAMPLE_BOUND) { - return 1; - } - - return 0; } /** @@ -1083,6 +1066,9 @@ static void update_sub_state(AMRWBContext *ctx) memmove(&ctx->pitch_gain[0], &ctx->pitch_gain[1], 4 * sizeof(float)); memmove(&ctx->fixed_gain[0], &ctx->fixed_gain[1], 4 * sizeof(float)); + + memmove(&ctx->samples_in[0], &ctx->samples_in[AMRWB_SUBFRAME_SIZE], + LP_ORDER * sizeof(float)); } static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, @@ -1174,7 +1160,6 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, } /* Post-processing of excitation elements */ - /* XXX: Noise_enhancer was not tested yet, needs that 1.5dB check */ synth_fixed_gain = noise_enhancer(ctx->fixed_gain[4], &ctx->prev_tr_gain, voice_fac, stab_fac); @@ -1186,13 +1171,8 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, pitch_enhancer(synth_fixed_vector, voice_fac); - if (synthesis(ctx, ctx->lp_coef[sub], synth_exc, synth_fixed_gain, - synth_fixed_vector, &ctx->samples_in[LP_ORDER], 0)) - // overflow detected -> rerun synthesis scaling pitch vector down - // by a factor of 4, skipping pitch vector contribution emphasis - // and adaptive gain control - synthesis(ctx, ctx->lp_coef[sub], synth_exc, synth_fixed_gain, - synth_fixed_vector, &ctx->samples_in[LP_ORDER], 1); + synthesis(ctx, ctx->lp_coef[sub], synth_exc, synth_fixed_gain, + synth_fixed_vector, &ctx->samples_in[LP_ORDER]); /* Synthesis speech post-processing */ de_emphasis(&ctx->samples_in[LP_ORDER], PREEMPH_FAC, ctx->demph_mem); commit fe78e3f003f346c235c38403442d8f4f9f541afd Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Mon Jul 26 17:57:28 2010 -0300 Change the stability factor multiplier to 512, accurate according to some tests diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 208c7c4..8768863 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -78,7 +78,7 @@ typedef struct { float hpf_31_mem[4], hpf_400_mem[4]; ///< previous values in the high-pass filters AVLFG prng; ///< random number generator for white noise excitation - uint8_t first_frame; ///< flag active in the first frame decoded + uint8_t first_frame; ///< flag active during decoding of the first frame } AMRWBContext; static av_cold int amrwb_decode_init(AVCodecContext *avctx) @@ -437,7 +437,7 @@ static void decode_pitch_vector(AMRWBContext *ctx, LP_ORDER, AMRWB_SUBFRAME_SIZE); /* Check which pitch signal path should be used. - * 6k60 and 8k85 modes have ltp flag set to 0 */ + * 6k60 and 8k85 modes have the ltp flag set to 0 */ if (amr_subframe->ltp) { memcpy(ctx->pitch_vector, exc, AMRWB_SUBFRAME_SIZE * sizeof(float)); } else { @@ -805,8 +805,8 @@ static float stability_factor(const float *isf, const float *isf_past) acc += (isf[i] - isf_past[i]) * (isf[i] - isf_past[i]); // XXX: I could not understand well this part from ref code - // it made more sense changing the "/ 256" to "* 256" - return FFMAX(0.0, 1.25 - acc * 0.8 * 256); + // it made more sense changing the "/ 256" to "* 512" + return FFMAX(0.0, 1.25 - acc * 0.8 * 512); } /** @@ -855,6 +855,9 @@ static void pitch_enhancer(float *fixed_vector, float voice_fac) float cpe = 0.125 * (1 + voice_fac); float last = fixed_vector[0]; // holds c(i - 1) + /* XXX: This procedure seems correct, but due to some roundings + * in the opencore code (line 1037 onwards) the resulting fixed_vector + * differs quite a bit between the two implementations */ fixed_vector[0] -= cpe * fixed_vector[1]; for (i = 1; i < AMRWB_SUBFRAME_SIZE - 1; i++) { ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdata.h | 36 +++++++++++++++++-- libavcodec/amrwbdec.c | 95 +++++++++++++++++++++++++++--------------------- 2 files changed, 86 insertions(+), 45 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0

[soc] G.723.1 Decoder d5729e70838463f6b4a92b9e5914b15ce7d74b65
by naufal11＠gmail.com 26 Jul '10

26 Jul '10

- Log ----------------------------------------------------------------- commit d5729e70838463f6b4a92b9e5914b15ce7d74b65 Author: Naufal <naufal11(a)gmail.com> Date: Mon Jul 26 22:04:59 2010 +0530 Perform postfilter gain scaling diff --git a/libavcodec/g723_1.c b/libavcodec/g723_1.c index f15860f..d0beff3 100755 --- a/libavcodec/g723_1.c +++ b/libavcodec/g723_1.c @@ -27,6 +27,8 @@ typedef struct g723_1_context { int16_t reflection_coef; int16_t pf_fir_mem[LPC_ORDER]; ///< Formant FIR postfilter memory int pf_iir_mem[LPC_ORDER]; ///< Formant IIR postfilter memory + int16_t pf_gain; ///< Formant postfilter + ///< gain scaling unit memory } G723_1_Context; static av_cold int g723_1_decode_init(AVCodecContext *avctx) @@ -38,6 +40,7 @@ static av_cold int g723_1_decode_init(AVCodecContext *avctx) avctx->sample_rate = 8000; memcpy(p->prev_lsp, dc_lsp, LPC_ORDER * sizeof(int16_t)); + p->pf_gain = 1 << 12; return 0; } @@ -777,6 +780,47 @@ static int formant_postfilter(G723_1_Context *p, int16_t *buf, int16_t *lpc) return energy; } +/* + * Adjust gain of postfiltered signal. + * + * @param buf postfiltered output vector + * @param energy input energy coefficient + */ +static void gain_scale(G723_1_Context *p, int16_t * buf, int energy) +{ + int num, denom, gain, bits1, bits2; + int temp, i; + + num = energy; + denom = 0; + for (i = 0; i < SUBFRAME_LEN; i++) { + temp = av_clipl_int32((int64_t)(buf[i] >> 2) * (buf[i] >> 2) << 1); + denom = av_clipl_int32((int64_t)denom + temp); + } + + if (num && denom) { + bits1 = normalize_bits_int32(num); + bits2 = normalize_bits_int32(denom); + num = num << bits1 >> 1; + denom <<= bits2; + + bits2 = 5 + bits1 - bits2; + bits2 = FFMAX(0, bits2); + + gain = (num >> 1) / (denom >> 16); + gain = gain << 16 >> bits2; + gain = ff_sqrt(gain >> 1); + } else { + gain = 1 << 12; + } + + for (i = 0; i < SUBFRAME_LEN; i++) { + p->pf_gain = ((p->pf_gain << 4) - p->pf_gain + gain + (1 << 3)) >> 4; + buf[i] = av_clip_int16((buf[i] * (p->pf_gain + (p->pf_gain >> 4)) + + (1 << 10)) >> 11); + } +} + static int g723_1_decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt) { @@ -888,6 +932,7 @@ static int g723_1_decode_frame(AVCodecContext *avctx, void *data, LPC_ORDER * sizeof(int16_t)); energy = formant_postfilter(p, vector_ptr, &lpc[i * (LPC_ORDER + 1) + 1]); + gain_scale(p, vector_ptr, energy); vector_ptr += SUBFRAME_LEN; } ----------------------------------------------------------------------- Summary of changes: libavcodec/g723_1.c | 45 +++++++++++++++++++++++++++++++++++++++++++++ 1 files changed, 45 insertions(+), 0 deletions(-) -- http://github.com/naufal/ffmpeg-soc

1 0

[soc] G.723.1 Decoder a7b84565a34cc81ceb2d60bc871a95eaafea6474
by naufal11＠gmail.com 26 Jul '10

26 Jul '10

- Log ----------------------------------------------------------------- commit a7b84565a34cc81ceb2d60bc871a95eaafea6474 Author: Naufal <naufal11(a)gmail.com> Date: Mon Jul 26 20:41:09 2010 +0530 Add formant postfilter diff --git a/libavcodec/g723_1.c b/libavcodec/g723_1.c index 68653e5..f15860f 100755 --- a/libavcodec/g723_1.c +++ b/libavcodec/g723_1.c @@ -23,6 +23,10 @@ typedef struct g723_1_context { int16_t interp_gain; int16_t sid_gain; int16_t cur_gain; + + int16_t reflection_coef; + int16_t pf_fir_mem[LPC_ORDER]; ///< Formant FIR postfilter memory + int pf_iir_mem[LPC_ORDER]; ///< Formant IIR postfilter memory } G723_1_Context; static av_cold int g723_1_decode_init(AVCodecContext *avctx) @@ -686,6 +690,93 @@ static void residual_interp(int16_t *buf, int16_t *out, int16_t lag, } } +/* + * @param filter_coef filter coefficients vector + * @param src input signal vector + * @param dest output filtered vector + */ +static void iir_filter(int16_t filter_coef[][LPC_ORDER], int16_t *src, + int *dest) +{ + int i, j; + + for (i = 0; i < SUBFRAME_LEN; i++) { + int64_t filter = 0; + + for (j = 1; j <= LPC_ORDER; j++) { + filter -= filter_coef[0][j - 1] * src[i - j] - + filter_coef[1][j - 1] * (dest[i - j] >> 16); + } + + dest[i] = av_clipl_int32((src[i] << 16) + (filter << 3) + (1 << 15)); + } +} + +/* + * @param buf synthesized speech vector + * @param lpc quantized lpc coefficients + */ +static int formant_postfilter(G723_1_Context *p, int16_t *buf, int16_t *lpc) +{ + int16_t filter_coef[2][LPC_ORDER]; + int filter_signal[LPC_ORDER+SUBFRAME_LEN]; + int16_t temp_vector[SUBFRAME_LEN]; + int auto_corr[2]; + int16_t temp; + int scale, energy; + int i; + + // Compute FIR and IIR filter coefficients + for (i=0; i<LPC_ORDER; i++) { + filter_coef[0][i] = (-lpc[i] * postfilter_tbl[0][i] + (1<<14)) >> 15; + filter_coef[1][i] = (-lpc[i] * postfilter_tbl[1][i] + (1<<14)) >> 15; + } + + // Normalize + memcpy(temp_vector, buf, SUBFRAME_LEN * sizeof(int16_t)); + scale = scale_vector(temp_vector, SUBFRAME_LEN); + + // Compute auto correlation coefficients + auto_corr[0] = dot_product(temp_vector, temp_vector + 1, SUBFRAME_LEN - 1); + auto_corr[1] = dot_product(temp_vector, temp_vector, SUBFRAME_LEN); + + // Compute normalized signal energy + temp = 2 * scale + 4; + if (temp < 0) + energy = av_clipl_int32((int64_t)auto_corr[1] << -temp); + else + energy = auto_corr[1] >> temp; + + // Compute reflection coefficient + temp = auto_corr[1] >> 16; + if (temp) { + temp = (auto_corr[0]>>2) / temp; + } + + p->reflection_coef = ((p->reflection_coef << 2) - p->reflection_coef + + temp + 2) >> 2; + + memcpy(buf - LPC_ORDER, p->pf_fir_mem, LPC_ORDER * sizeof(int16_t)); + memcpy(filter_signal, p->pf_iir_mem, LPC_ORDER * sizeof(int)); + + iir_filter(filter_coef, buf, filter_signal + LPC_ORDER); + + memcpy(p->pf_fir_mem, buf + SUBFRAME_LEN - LPC_ORDER, LPC_ORDER * + sizeof(int16_t)); + memcpy(p->pf_iir_mem, filter_signal + SUBFRAME_LEN, LPC_ORDER * + sizeof(int)); + + // Compensation filter + temp = (p->reflection_coef * 0xffffc000 >> 15) & 0xfffc; + for (i = 0; i < SUBFRAME_LEN; i++) { + buf[i] = av_clipl_int32(filter_signal[LPC_ORDER + i] + + (((filter_signal[LPC_ORDER + i - 1] >> 16) * + temp) << 1)) >> 16; + } + + return energy; +} + static int g723_1_decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPacket *avpkt) { @@ -700,6 +791,7 @@ static int g723_1_decode_frame(AVCodecContext *avctx, void *data, int16_t excitation[FRAME_LEN + PITCH_MAX]; int16_t acb_vector[SUBFRAME_LEN]; int16_t *vector_ptr; + int energy; int bad_frame = 0, i, j; if (!buf_size || buf_size < frame_size[buf[0] & 3]) { @@ -794,6 +886,8 @@ static int g723_1_decode_frame(AVCodecContext *avctx, void *data, // Store the last 10 values of the output vector memcpy(p->filter_mem, vector_ptr + SUBFRAME_LEN - LPC_ORDER, LPC_ORDER * sizeof(int16_t)); + energy = formant_postfilter(p, vector_ptr, + &lpc[i * (LPC_ORDER + 1) + 1]); vector_ptr += SUBFRAME_LEN; } diff --git a/libavcodec/g723_1_data.h b/libavcodec/g723_1_data.h index 411e23c..ec93acf 100644 --- a/libavcodec/g723_1_data.h +++ b/libavcodec/g723_1_data.h @@ -1206,3 +1206,13 @@ static const int16_t adaptive_cb_gain170[170 * 20] = { 6538, -1712, -4280, -4804, -3613, -2609, -2707, -2868, -4534, -2487, -3932, -4166, -2113, -3341, -3540, -3070 }; + +/* + * 0.65^i (FIR) and 0.75^i (IIR) scaled by 2^15 + */ +static const int16_t postfilter_tbl[2][LPC_ORDER] = { + // Zero part + {21299, 13844, 8999, 5849, 3802, 2471, 1606, 1044, 679, 441}, + // Pole part + {24576, 18432, 13824, 10368, 7776, 5832, 4374, 3281, 2460, 1845} +}; ----------------------------------------------------------------------- Summary of changes: libavcodec/g723_1.c | 94 ++++++++++++++++++++++++++++++++++++++++++++++ libavcodec/g723_1_data.h | 10 +++++ 2 files changed, 104 insertions(+), 0 deletions(-) -- http://github.com/naufal/ffmpeg-soc

1 0

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 25 Jul '10

25 Jul '10

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "AMR-WB decoder". The branch, master has been updated via 0ca8fed46c9ff9b50f5d51f2d1fa79d8583aedab (commit) via dae60194c8f5fa005f3e74c9632c8f624b4ee33e (commit) via 7d21ff9007c7720896ca40a96cc1db5d4b5c7b62 (commit) from 97cc97cb5735e660655029a0820fb8baf7fbd256 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit 0ca8fed46c9ff9b50f5d51f2d1fa79d8583aedab Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Sun Jul 25 00:53:41 2010 -0300 Improve precision in the resulting fixed vector from anti-sparseness processing by adapting the pitch lag diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index a5fae3d..208c7c4 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -425,7 +425,7 @@ static void decode_pitch_vector(AMRWBContext *ctx, decode_pitch_lag_high(&pitch_lag_int, &pitch_lag_frac, amr_subframe->adap, &ctx->base_pitch_lag, subframe); - ctx->pitch_lag_int = pitch_lag_int; + ctx->pitch_lag_int = pitch_lag_int + (pitch_lag_frac < 0 ? -1 : 0); pitch_lag_int += pitch_lag_frac > 0; /* Calculate the pitch vector by interpolating the past excitation at the @@ -681,9 +681,6 @@ static void decode_gains(const uint8_t vq_gain, const enum Mode mode, static void pitch_sharpening(AMRWBContext *ctx, AMRFixed *fixed_sparse, float *fixed_vector) { - /* XXX: The resulting fixed_vector is a bit different comparing - * to opencore, probably because the filters are applied in reverse order */ - /* Periodicity enhancement part */ fixed_sparse->pitch_lag = ctx->pitch_lag_int; fixed_sparse->pitch_fac = 0.85; @@ -1149,7 +1146,6 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, decode_gains(cur_subframe->vq_gain, ctx->fr_cur_mode, &fixed_gain_factor, &ctx->pitch_gain[4]); - /* XXX: not tested yet */ pitch_sharpening(ctx, &fixed_sparse, ctx->fixed_vector); ctx->fixed_gain[4] = @@ -1164,8 +1160,6 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, ctx->fixed_vector, ctx->fixed_gain[4]); ctx->tilt_coef = voice_fac * 0.25 + 0.25; - /* XXX: Tested against the ref code until here */ - /* Construct current excitation */ for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) { ctx->excitation[i] *= ctx->pitch_gain[4]; @@ -1177,12 +1171,16 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, } /* Post-processing of excitation elements */ + /* XXX: Noise_enhancer was not tested yet, needs that 1.5dB check */ synth_fixed_gain = noise_enhancer(ctx->fixed_gain[4], &ctx->prev_tr_gain, voice_fac, stab_fac); synth_fixed_vector = anti_sparseness(ctx, ctx->fixed_vector, spare_vector); + /* XXX: Tested against the ref code until here, it "succeeds" at least + * for cases in which the "opencore bug" don't interfere */ + pitch_enhancer(synth_fixed_vector, voice_fac); if (synthesis(ctx, ctx->lp_coef[sub], synth_exc, synth_fixed_gain, commit dae60194c8f5fa005f3e74c9632c8f624b4ee33e Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Sun Jul 25 00:48:37 2010 -0300 Change reordering tables to parse individual tracks in the 6k60 mode pulse indexing diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 93e76d5..d8a1068 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -97,20 +97,20 @@ static const uint16_t order_MODE_6k60[] = { 7, AMR_BIT(isp_id[3]), 59, 60, 61, 62, 63, 76, 77, 6, AMR_BIT(isp_id[4]), 78, 79, 81, 80, 82, 83, 8, AMR_OF(0, adap), 12, 13, 14, 15, 27, 28, 54, 66, - 12, AMR_OF(0, pul_il[0]), 84, 92, 100, 108, 116, 124, 85, 96, - 104, 112, 120, 128, + 6, AMR_OF(0, pul_il[0]), 84, 92, 100, 108, 116, 124, + 6, AMR_OF(0, pul_il[1]), 85, 96, 104, 112, 120, 128, 6, AMR_OF(0, vq_gain), 33, 16, 37, 20, 4, 8, 5, AMR_OF(1, adap), 39, 46, 56, 68, 74, - 12, AMR_OF(1, pul_il[0]), 86, 93, 101, 109, 117, 125, 87, 97, - 105, 113, 121, 129, + 6, AMR_OF(1, pul_il[0]), 86, 93, 101, 109, 117, 125, + 6, AMR_OF(1, pul_il[1]), 87, 97, 105, 113, 121, 129, 6, AMR_OF(1, vq_gain), 29, 17, 35, 21, 5, 9, 5, AMR_OF(2, adap), 42, 47, 53, 55, 67, - 12, AMR_OF(2, pul_il[0]), 88, 94, 102, 110, 118, 126, 89, 98, - 106, 114, 122, 130, + 6, AMR_OF(2, pul_il[0]), 88, 94, 102, 110, 118, 126, + 6, AMR_OF(2, pul_il[1]), 89, 98, 106, 114, 122, 130, 6, AMR_OF(2, vq_gain), 34, 18, 36, 22, 6, 24, 5, AMR_OF(3, adap), 40, 41, 57, 69, 75, - 12, AMR_OF(3, pul_il[0]), 90, 95, 103, 111, 119, 127, 91, 99, - 107, 115, 123, 131, + 6, AMR_OF(3, pul_il[0]), 90, 95, 103, 111, 119, 127, + 6, AMR_OF(3, pul_il[1]), 91, 99, 107, 115, 123, 131, 6, AMR_OF(3, vq_gain), 30, 19, 26, 23, 7, 25, 0 }; commit 7d21ff9007c7720896ca40a96cc1db5d4b5c7b62 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Sat Jul 24 21:23:46 2010 -0300 Fix a bug in decoded pitch lag scale for low modes and remember not to apply anti_sparseness in high modes diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 0fb6da7..a5fae3d 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -404,7 +404,7 @@ static void decode_pitch_lag_low(int *lag_int, int *lag_frac, int pitch_index, AMRWB_P_DELAY_MAX - 15); } else { *lag_int = (pitch_index + 1) >> 1; - *lag_frac = pitch_index - (*lag_int << 1); + *lag_frac = (pitch_index - (*lag_int << 1)) << 1; *lag_int += *base_lag_int; } } @@ -681,6 +681,9 @@ static void decode_gains(const uint8_t vq_gain, const enum Mode mode, static void pitch_sharpening(AMRWBContext *ctx, AMRFixed *fixed_sparse, float *fixed_vector) { + /* XXX: The resulting fixed_vector is a bit different comparing + * to opencore, probably because the filters are applied in reverse order */ + /* Periodicity enhancement part */ fixed_sparse->pitch_lag = ctx->pitch_lag_int; fixed_sparse->pitch_fac = 0.85; @@ -725,6 +728,9 @@ static float *anti_sparseness(AMRWBContext *ctx, { int ir_filter_nr; + if (ctx->fr_cur_mode > MODE_8k85) // no filtering in high modes + return fixed_vector; + if (ctx->pitch_gain[4] < 0.6) { ir_filter_nr = 0; // strong filtering } else if (ctx->pitch_gain[4] < 0.9) { ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdata.h | 16 ++++++++-------- libavcodec/amrwbdec.c | 14 +++++++++----- 2 files changed, 17 insertions(+), 13 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 23 Jul '10

23 Jul '10

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "AMR-WB decoder". The branch, master has been updated via 97cc97cb5735e660655029a0820fb8baf7fbd256 (commit) via 112b01c4d00b126a982b3ce83884ad13ffa46343 (commit) from 30af71e0fb3351656d148d939ba401c980304d4f (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit 97cc97cb5735e660655029a0820fb8baf7fbd256 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Fri Jul 23 22:32:49 2010 -0300 Test pulses decoding, fix typo in decode_5p_track, update excitation for next subframe diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 489f9ad..0fb6da7 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -430,12 +430,14 @@ static void decode_pitch_vector(AMRWBContext *ctx, /* Calculate the pitch vector by interpolating the past excitation at the pitch lag using a hamming windowed sinc function. */ + /* XXX: Not tested yet, need to ensure correct excitation construction before */ ff_acelp_interpolatef(exc, exc + 1 - pitch_lag_int, ac_inter, 4, pitch_lag_frac + 4 - 4*(pitch_lag_frac > 0), LP_ORDER, AMRWB_SUBFRAME_SIZE); - /* Check which pitch signal path should be used */ + /* Check which pitch signal path should be used. + * 6k60 and 8k85 modes have ltp flag set to 0 */ if (amr_subframe->ltp) { memcpy(ctx->pitch_vector, exc, AMRWB_SUBFRAME_SIZE * sizeof(float)); } else { @@ -526,9 +528,9 @@ static void decode_5p_track(int *out, int code, int m, int off) { ///code: 5m bits int half_3p = BIT_POS(code, 5*m - 1) << (m - 1); - decode_3p_track(out, BIT_STR(code, 2*m, 3*m - 2), + decode_3p_track(out, BIT_STR(code, 2*m + 1, 3*m - 2), m - 1, off + half_3p); - // XXX: there seems to be a typo in I3p expoent (from reference) + decode_2p_track(out + 3, BIT_STR(code, 0, 2*m + 1), m, off); } @@ -1070,6 +1072,9 @@ static void extrapolate_isf(float *out, float *isf) */ static void update_sub_state(AMRWBContext *ctx) { + memmove(&ctx->excitation_buf[0], &ctx->excitation_buf[AMRWB_SUBFRAME_SIZE], + (AMRWB_P_DELAY_MAX + LP_ORDER + 1) * sizeof(float)); + memmove(&ctx->pitch_gain[0], &ctx->pitch_gain[1], 4 * sizeof(float)); memmove(&ctx->fixed_gain[0], &ctx->fixed_gain[1], 4 * sizeof(float)); } @@ -1126,8 +1131,6 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, for (sub = 0; sub < 4; sub++) isp2lp(ctx->isp[sub], ctx->lp_coef[sub], LP_ORDER/2); - /* XXX: Tested against the ref code until here */ - for (sub = 0; sub < 4; sub++) { const AMRWBSubFrame *cur_subframe = &cf->subframe[sub]; @@ -1140,6 +1143,7 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, decode_gains(cur_subframe->vq_gain, ctx->fr_cur_mode, &fixed_gain_factor, &ctx->pitch_gain[4]); + /* XXX: not tested yet */ pitch_sharpening(ctx, &fixed_sparse, ctx->fixed_vector); ctx->fixed_gain[4] = @@ -1154,6 +1158,8 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, ctx->fixed_vector, ctx->fixed_gain[4]); ctx->tilt_coef = voice_fac * 0.25 + 0.25; + /* XXX: Tested against the ref code until here */ + /* Construct current excitation */ for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) { ctx->excitation[i] *= ctx->pitch_gain[4]; commit 112b01c4d00b126a982b3ce83884ad13ffa46343 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Fri Jul 23 15:55:55 2010 -0300 Pitch lag testing: bound past lag, unify 6k60 and 8k85 decodings, need to check round versus floor usage diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 7f221e4..93e76d5 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -34,7 +34,9 @@ #define AMRWB_SUBFRAME_SIZE 64 ///< samples per subframe at 12.8 kHz #define AMRWB_SFR_SIZE_OUT 80 ///< samples per subframe at 16 kHz #define AMRWB_SAMPLE_BOUND 32768.0 ///< threshold for synthesis overflow -#define PITCH_MAX 231 ///< maximum received pitch delay value + +#define AMRWB_P_DELAY_MAX 231 ///< maximum pitch delay value +#define AMRWB_P_DELAY_MIN 34 #define PREEMPH_FAC 0.68 ///< factor used to de-emphasize synthesis diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 4f0e765..489f9ad 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -56,7 +56,7 @@ typedef struct { uint8_t base_pitch_lag; ///< integer part of pitch lag for next relative subframe uint8_t pitch_lag_int; ///< integer part of pitch lag of the previous subframe - float excitation_buf[PITCH_MAX + LP_ORDER + 1 + AMRWB_SUBFRAME_SIZE]; ///< current excitation and all necessary excitation history + float excitation_buf[AMRWB_P_DELAY_MAX + LP_ORDER + 1 + AMRWB_SUBFRAME_SIZE]; ///< current excitation and all necessary excitation history float *excitation; ///< points to current excitation in excitation_buf[] float pitch_vector[AMRWB_SUBFRAME_SIZE]; ///< adaptive codebook (pitch) vector for current subframe @@ -90,7 +90,7 @@ static av_cold int amrwb_decode_init(AVCodecContext *avctx) av_lfg_init(&ctx->prng, 1); - ctx->excitation = &ctx->excitation_buf[PITCH_MAX + LP_ORDER + 1]; + ctx->excitation = &ctx->excitation_buf[AMRWB_P_DELAY_MAX + LP_ORDER + 1]; ctx->first_frame = 1; ctx->tilt_coef = ctx->prev_tr_gain = 0.0; @@ -254,11 +254,11 @@ static void decode_isf_indices_46b(uint16_t *ind, float *isf_q, uint8_t fr_q) { * Apply mean and past ISF values using the prediction factor * Updates past ISF vector * - * @param isf_q [in] current quantized ISF + * @param isf_q [in/out] current quantized ISF * @param isf_past [in/out] past quantized ISF * */ -static void isf_add_mean_and_past(const float *isf_q, float *isf_past) { +static void isf_add_mean_and_past(float *isf_q, float *isf_past) { int i; float tmp; @@ -295,7 +295,7 @@ static void isf_set_min_dist(float *isf, float min_spacing, int size) { * @param isp_q [in/out] ISPs for each subframe * @param isp4_past [in] Past ISP for subframe 4 */ -static void interpolate_isp(double *isp_q[4], const double *isp4_past) +static void interpolate_isp(double isp_q[4][LP_ORDER], const double *isp4_past) { int i; // XXX: Did not use ff_weighted_vector_sumf because using double @@ -355,7 +355,7 @@ static void isp2lp(const double *isp, float *lp, int lp_half_order) { * @param subframe [in] Current subframe index (0 to 3) */ static void decode_pitch_lag_high(int *lag_int, int *lag_frac, int pitch_index, - uint8_t *base_lag_int, const int subframe) + uint8_t *base_lag_int, int subframe) { if (subframe == 0 || subframe == 2) { if (pitch_index < 376) { @@ -369,47 +369,30 @@ static void decode_pitch_lag_high(int *lag_int, int *lag_frac, int pitch_index, *lag_int = pitch_index - 280; *lag_frac = 0; } - *base_lag_int = *lag_int; // store previous lag + /* minimum lag for next subframe */ + *base_lag_int = av_clip(*lag_int - 8, AMRWB_P_DELAY_MIN, + AMRWB_P_DELAY_MAX - 15); + /* XXX: the spec states clearly that *base_lag_int should be + * the nearest integer to *lag_int (minus 8), but the ref code + * actually uses always its floor, causing the next frame integer + * lag to be one less than mine when the nearest integer is + * not equal to the floor */ } else { *lag_int = (pitch_index + 1) >> 2; *lag_frac = pitch_index - (*lag_int << 2); - *lag_int += *base_lag_int - 8; - // XXX: Doesn't seem to need bounding according to TS 26.190 + *lag_int += *base_lag_int; } } /** - * Decode a adaptive codebook index into pitch lag for 8k85 mode - * Description is analogous to decode_pitch_lag_high - */ -static void decode_pitch_lag_8K85(int *lag_int, int *lag_frac, int pitch_index, - uint8_t *base_lag_int, const int subframe) -{ - if (subframe == 0 || subframe == 2) { - if (pitch_index < 116) { - *lag_int = (pitch_index + 69) >> 1; - *lag_frac = (pitch_index - (*lag_int << 1) + 68) << 1; - } else { - *lag_int = pitch_index - 24; - *lag_frac = 0; - } - *base_lag_int = *lag_int; - } else { - *lag_int = (pitch_index + 1) >> 1; - *lag_frac = pitch_index - (*lag_int << 1); - *lag_int += *base_lag_int - 8; - } -} - -/** - * Decode a adaptive codebook index into pitch lag for 6k60 mode - * Description is analogous to decode_pitch_lag_high, but relative + * Decode a adaptive codebook index into pitch lag for 8k85 and 6k60 modes + * Description is analogous to decode_pitch_lag_high, but in 6k60 relative * index is used for all subframes except the first */ -static void decode_pitch_lag_6K60(int *lag_int, int *lag_frac, int pitch_index, - uint8_t *base_lag_int, const int subframe) +static void decode_pitch_lag_low(int *lag_int, int *lag_frac, int pitch_index, + uint8_t *base_lag_int, int subframe, enum Mode mode) { - if (subframe == 0) { + if (subframe == 0 || (subframe == 2 && mode != MODE_6k60)) { if (pitch_index < 116) { *lag_int = (pitch_index + 69) >> 1; *lag_frac = (pitch_index - (*lag_int << 1) + 68) << 1; @@ -417,11 +400,12 @@ static void decode_pitch_lag_6K60(int *lag_int, int *lag_frac, int pitch_index, *lag_int = pitch_index - 24; *lag_frac = 0; } - *base_lag_int = *lag_int; + *base_lag_int = av_clip(*lag_int - 8, AMRWB_P_DELAY_MIN, + AMRWB_P_DELAY_MAX - 15); } else { *lag_int = (pitch_index + 1) >> 1; *lag_frac = pitch_index - (*lag_int << 1); - *lag_int += *base_lag_int - 8; + *lag_int += *base_lag_int; } } @@ -434,12 +418,9 @@ static void decode_pitch_vector(AMRWBContext *ctx, float *exc = ctx->excitation; enum Mode mode = ctx->fr_cur_mode; - if (mode == MODE_6k60) { - decode_pitch_lag_6K60(&pitch_lag_int, &pitch_lag_frac, amr_subframe->adap, - &ctx->base_pitch_lag, subframe); - } else if (mode == MODE_8k85) { - decode_pitch_lag_8K85(&pitch_lag_int, &pitch_lag_frac, amr_subframe->adap, - &ctx->base_pitch_lag, subframe); + if (mode <= MODE_8k85) { + decode_pitch_lag_low(&pitch_lag_int, &pitch_lag_frac, amr_subframe->adap, + &ctx->base_pitch_lag, subframe, mode); } else decode_pitch_lag_high(&pitch_lag_int, &pitch_lag_frac, amr_subframe->adap, &ctx->base_pitch_lag, subframe); @@ -1010,9 +991,8 @@ static float find_hb_gain(AMRWBContext *ctx, const float *synth, ff_dot_productf(synth, synth, AMRWB_SUBFRAME_SIZE); tilt = FFMAX(0.0, tilt); - /* return gain bounded by [0.1, 1.0] */ - return FFMAX(0.1, FFMIN(1.0, (1.0 - tilt) * (1.25 - 0.25 * wsp))); + return av_clipf((1.0 - tilt) * (1.25 - 0.25 * wsp), 0.1, 1.0); } /** ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdata.h | 4 ++- libavcodec/amrwbdec.c | 90 ++++++++++++++++++++--------------------------- 2 files changed, 41 insertions(+), 53 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 23 Jul '10

23 Jul '10

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "AMR-WB decoder". The branch, master has been updated via 30af71e0fb3351656d148d939ba401c980304d4f (commit) from ade74ca125016eab5a8c5766e192bd8885686b01 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit 30af71e0fb3351656d148d939ba401c980304d4f Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Fri Jul 23 01:15:27 2010 -0300 Various small fixes to the ISP/LP vector calculation, added "const" to some ISF/ISP function headers diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index a0bf66d..4f0e765 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -176,8 +176,10 @@ static void isf2isp(const float *isf, double *isp) { int i; - for (i = 0; i < LP_ORDER; i++) + for (i = 0; i < LP_ORDER - 1; i++) isp[i] = cos(2.0 * M_PI * isf[i]); + + isp[LP_ORDER - 1] = cos(4.0 * M_PI * isf[LP_ORDER - 1]); } /** @@ -256,7 +258,7 @@ static void decode_isf_indices_46b(uint16_t *ind, float *isf_q, uint8_t fr_q) { * @param isf_past [in/out] past quantized ISF * */ -static void isf_add_mean_and_past(float *isf_q, float *isf_past) { +static void isf_add_mean_and_past(const float *isf_q, float *isf_past) { int i; float tmp; @@ -293,7 +295,7 @@ static void isf_set_min_dist(float *isf, float min_spacing, int size) { * @param isp_q [in/out] ISPs for each subframe * @param isp4_past [in] Past ISP for subframe 4 */ -static void interpolate_isp(double isp_q[4][LP_ORDER], double *isp4_past) +static void interpolate_isp(double *isp_q[4], const double *isp4_past) { int i; // XXX: Did not use ff_weighted_vector_sumf because using double @@ -316,28 +318,29 @@ static void interpolate_isp(double isp_q[4][LP_ORDER], double *isp4_past) * @param lp [out] LP coefficients * @param lp_half_order [in] Half the number of LPs to construct */ -static void isp2lp(double isp[LP_ORDER], float *lp, int lp_half_order) { +static void isp2lp(const double *isp, float *lp, int lp_half_order) { double pa[MAX_LP_HALF_ORDER+1], qa[MAX_LP_HALF_ORDER+1]; float *lp2 = lp + (lp_half_order << 1); double last_isp = isp[2 * lp_half_order - 1]; - double qa_old = 0; // XXX: qa[i-2] assuming qa[-1] = 0, not mentioned in spec + double qa_old = 0.0; int i; ff_lsp2polyf(isp, pa, lp_half_order); - ff_lsp2polyf(isp + 1, qa, lp_half_order); + ff_lsp2polyf(isp + 1, qa, lp_half_order - 1); - for (i=1; i<lp_half_order; i++) { + for (i = 1; i < lp_half_order; i++) { double paf = (1 + last_isp) * pa[i]; double qaf = (1 - last_isp) * (qa[i] - qa_old); - qa_old = qa[i-1]; + qa_old = qa[i - 1]; - lp[i] = 0.5 * (paf + qaf); - lp2[i] = 0.5 * (paf - qaf); + lp[i] = 0.5 * (paf + qaf); + lp2[-i] = 0.5 * (paf - qaf); } - lp2[0] = 0.5 * (1 + last_isp) * pa[lp_half_order] * lp_half_order; - lp2[lp_half_order] = last_isp; + lp[0] = 1.0; + lp[lp_half_order] = 0.5 * (1 + last_isp) * pa[lp_half_order]; + lp2[0] = last_isp; } /** @@ -1140,11 +1143,11 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, } interpolate_isp(ctx->isp, ctx->isp_sub4_past); - /* XXX: Tested against the ref code until here */ - for (sub = 0; sub < 4; sub++) isp2lp(ctx->isp[sub], ctx->lp_coef[sub], LP_ORDER/2); + /* XXX: Tested against the ref code until here */ + for (sub = 0; sub < 4; sub++) { const AMRWBSubFrame *cur_subframe = &cf->subframe[sub]; ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdec.c | 31 +++++++++++++++++-------------- 1 files changed, 17 insertions(+), 14 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 22 Jul '10

22 Jul '10

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "AMR-WB decoder". The branch, master has been updated via ade74ca125016eab5a8c5766e192bd8885686b01 (commit) via ee327e3dbb0cfa189db2b12dce8ea9154ade2943 (commit) via 5216a6b9e872b5c23c56f91011a83ac87c19b329 (commit) via 50f3a13500de12d7d9fb3b382965c133a9eea200 (commit) from c2a8e2ad2d9c51c9d1f7b9101175db05f2fecabb (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit ade74ca125016eab5a8c5766e192bd8885686b01 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Thu Jul 22 02:35:50 2010 -0300 Sketch the LP filter (6k60 mode) for the high freq band, up to the autocorrelation calculation diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index d16253a..7f221e4 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -25,6 +25,7 @@ #include <stdint.h> #define LP_ORDER 16 ///< linear predictive coding filter order +#define LP_ORDER_16k 20 ///< lpc filter order at 16kHz #define MIN_ISF_SPACING (128 / 32768.0) ///< minimum isf gap #define PRED_FACTOR (1.0 / 3.0) #define MIN_ENERGY -14.0 ///< initial innnovation energy (dB) diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 58bbff6..a0bf66d 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -1038,6 +1038,50 @@ static void scaled_hb_excitation(AMRWBContext *ctx, float *hb_exc, hb_exc[i] *= hb_gain; } +static float auto_correlation(float *diff_isf, float mean, int lag) +{ + int i; + float sum = 0.0; + + for (i = 7; i < LP_ORDER - 2; i++) { + float prod = (diff_isf[i] - mean) * (diff_isf[i - lag] - mean); + sum += prod * prod; + } + return sum; +} + +static void extrapolate_isf(float *out, float *isf) +{ + float diff_isf[LP_ORDER - 2], diff_mean; + float corr_lag[3]; + int i, i_max_corr; + + memcpy(out, isf, LP_ORDER - 1); + out[LP_ORDER_16k - 1] = isf[LP_ORDER - 1]; + + /* Calculate the difference vector */ + for (i = 0; i < LP_ORDER - 2; i++) + diff_isf[i] = isf[i + 1] - isf[i]; + + diff_mean = 0.0; + for (i = 2; i < LP_ORDER - 2; i++) + diff_mean += diff_isf[i] / (LP_ORDER - 4); + + /* Find which is the maximum autocorrelation */ + i_max_corr = 0; + for (i = 0; i < 3; i++) { + corr_lag[i] = auto_correlation(diff_isf, diff_mean, i + 2); + if (corr_lag[i] > corr_lag[i_max_corr]) + i_max_corr = i; + } + i_max_corr++; + + for (i = LP_ORDER - 1; i < LP_ORDER_16k - 1; i++) + out[i] = isf[i - 1] + isf[i - 1 - i_max_corr] + - isf[i - 2 - i_max_corr]; + return; +} + /** * Update context state before the next subframe */ commit ee327e3dbb0cfa189db2b12dce8ea9154ade2943 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Thu Jul 22 02:33:20 2010 -0300 Test: Fix ISF/ISP initialization, add new past ISF, fix typo in isf_set_min_dist, add ISP special case when in the first frame diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index a4404be..d16253a 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -25,8 +25,8 @@ #include <stdint.h> #define LP_ORDER 16 ///< linear predictive coding filter order -#define MIN_ISF_SPACING 50.0 // XXX: Taken from fixed-point 26.173, not sure -#define PRED_FACTOR (1.0/3.0) +#define MIN_ISF_SPACING (128 / 32768.0) ///< minimum isf gap +#define PRED_FACTOR (1.0 / 3.0) #define MIN_ENERGY -14.0 ///< initial innnovation energy (dB) #define ENERGY_MEAN 30.0 ///< mean innovation energy (dB) in all modes @@ -1611,10 +1611,10 @@ static const int16_t isf_mean[LP_ORDER] = { 8750, 9753, 10705, 11728, 12833, 13971, 15043, 4037 }; -/* Initialization tables for the ISP vector */ -static const int16_t isp_init[LP_ORDER] = { - 32138, 30274, 27246, 23170, 18205, 12540, 6393, 0, - -6393, -12540, -18205, -23170, -27246, -30274, -32138, 1475 +/* Initialization tables for the processed ISF vector */ +static const int16_t isf_init[LP_ORDER] = { + 1024, 2048, 3072, 4096, 5120, 6144, 7168, 8192, + 9216, 10240, 11264, 12288, 13312, 14336, 15360, 3840 }; /* Coefficients for FIR interpolation of excitation vector diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 0281ae1..58bbff6 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -45,8 +45,9 @@ typedef struct { uint8_t fr_mode_ind; ///< mode indication field uint8_t fr_mode_req; ///< mode request field uint8_t fr_crc; ///< crc for class A bits - float isf_quant[LP_ORDER]; ///< quantized ISF vector from current frame + float isf_cur[LP_ORDER]; ///< working ISF vector from current frame float isf_q_past[LP_ORDER]; ///< quantized ISF vector of the previous frame + float isf_past_final[LP_ORDER]; ///< final processed ISF vector of the prev frame double isp[4][LP_ORDER]; ///< ISP vectors from current frame double isp_sub4_past[LP_ORDER]; ///< ISP vector for the 4th subframe of the previous frame @@ -77,6 +78,7 @@ typedef struct { float hpf_31_mem[4], hpf_400_mem[4]; ///< previous values in the high-pass filters AVLFG prng; ///< random number generator for white noise excitation + uint8_t first_frame; ///< flag active in the first frame decoded } AMRWBContext; static av_cold int amrwb_decode_init(AVCodecContext *avctx) @@ -88,12 +90,12 @@ static av_cold int amrwb_decode_init(AVCodecContext *avctx) av_lfg_init(&ctx->prng, 1); - ctx->excitation = &ctx->excitation_buf[PITCH_MAX + LP_ORDER + 1]; + ctx->excitation = &ctx->excitation_buf[PITCH_MAX + LP_ORDER + 1]; + ctx->first_frame = 1; + ctx->tilt_coef = ctx->prev_tr_gain = 0.0; for (i = 0; i < LP_ORDER; i++) - ctx->isp_sub4_past[i] = isp_init[i] / (float) (1 << 15); - - ctx->tilt_coef = ctx->prev_tr_gain = 0.0; + ctx->isf_past_final[i] = isf_init[i] / (float) (1 << 15); for (i = 0; i < 4; i++) ctx->prediction_error[i] = MIN_ENERGY; @@ -278,7 +280,7 @@ static void isf_set_min_dist(float *isf, float min_spacing, int size) { int i; float prev = 0.0; - for (i = 0; i < size; i++) { + for (i = 0; i < size - 1; i++) { isf[i] = FFMAX(isf[i], prev + min_spacing); prev = isf[i]; } @@ -813,7 +815,9 @@ static float stability_factor(const float *isf, const float *isf_past) for (i = 0; i < LP_ORDER - 1; i++) acc += (isf[i] - isf_past[i]) * (isf[i] - isf_past[i]); - return 1.25 - acc * 0.8 / 256; + // XXX: I could not understand well this part from ref code + // it made more sense changing the "/ 256" to "* 256" + return FFMAX(0.0, 1.25 - acc * 0.8 * 256); } /** @@ -1073,21 +1077,27 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, /* Decode the quantized ISF vector */ if (ctx->fr_cur_mode == MODE_6k60) { - decode_isf_indices_36b(cf->isp_id, ctx->isf_quant, ctx->fr_quality); + decode_isf_indices_36b(cf->isp_id, ctx->isf_cur, ctx->fr_quality); } else { - decode_isf_indices_46b(cf->isp_id, ctx->isf_quant, ctx->fr_quality); + decode_isf_indices_46b(cf->isp_id, ctx->isf_cur, ctx->fr_quality); } - stab_fac = stability_factor(ctx->isf_quant, ctx->isf_q_past); + isf_add_mean_and_past(ctx->isf_cur, ctx->isf_q_past); + isf_set_min_dist(ctx->isf_cur, MIN_ISF_SPACING, LP_ORDER); - isf_add_mean_and_past(ctx->isf_quant, ctx->isf_q_past); - isf_set_min_dist(ctx->isf_quant, MIN_ISF_SPACING, LP_ORDER); + stab_fac = stability_factor(ctx->isf_cur, ctx->isf_past_final); - isf2isp(ctx->isf_quant, ctx->isp[3]); + isf2isp(ctx->isf_cur, ctx->isp[3]); /* Generate a ISP vector for each subframe */ + if (ctx->first_frame) { + ctx->first_frame = 0; + memcpy(ctx->isp_sub4_past, ctx->isp[3], LP_ORDER * sizeof(double)); + } interpolate_isp(ctx->isp, ctx->isp_sub4_past); + /* XXX: Tested against the ref code until here */ + for (sub = 0; sub < 4; sub++) isp2lp(ctx->isp[sub], ctx->lp_coef[sub], LP_ORDER/2); @@ -1170,6 +1180,7 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, // update state for next frame memcpy(ctx->isp_sub4_past, ctx->isp[3], LP_ORDER * sizeof(ctx->isp[3][0])); + memcpy(ctx->isf_past_final, ctx->isf_cur, LP_ORDER * sizeof(float)); /* report how many samples we got */ *data_size = 4 * AMRWB_SFR_SIZE_OUT * sizeof(float); commit 5216a6b9e872b5c23c56f91011a83ac87c19b329 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Wed Jul 21 21:32:30 2010 -0300 Fix the ISF past vector initialization, now to zero diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 29624fa..a4404be 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -1611,12 +1611,7 @@ static const int16_t isf_mean[LP_ORDER] = { 8750, 9753, 10705, 11728, 12833, 13971, 15043, 4037 }; -/* Initialization tables for ISF and ISP vectors */ -static const int16_t isf_init[LP_ORDER] = { - 1024, 2048, 3072, 4096, 5120, 6144, 7168, 8192, - 9216, 10240, 11264, 12288, 13312, 14336, 15360, 3840 -}; - +/* Initialization tables for the ISP vector */ static const int16_t isp_init[LP_ORDER] = { 32138, 30274, 27246, 23170, 18205, 12540, 6393, 0, -6393, -12540, -18205, -23170, -27246, -30274, -32138, 1475 diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 992bcdc..0281ae1 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -90,10 +90,8 @@ static av_cold int amrwb_decode_init(AVCodecContext *avctx) ctx->excitation = &ctx->excitation_buf[PITCH_MAX + LP_ORDER + 1]; - for (i = 0; i < LP_ORDER; i++) { - ctx->isf_q_past[i] = isf_init[i] / (float) (1 << 15); + for (i = 0; i < LP_ORDER; i++) ctx->isp_sub4_past[i] = isp_init[i] / (float) (1 << 15); - } ctx->tilt_coef = ctx->prev_tr_gain = 0.0; @@ -262,8 +260,8 @@ static void isf_add_mean_and_past(float *isf_q, float *isf_past) { for (i = 0; i < LP_ORDER; i++) { tmp = isf_q[i]; - isf_q[i] = tmp + isf_mean[i] / (float) (1<<15); - isf_q[i] = isf_q[i] + PRED_FACTOR * isf_past[i]; + isf_q[i] += isf_mean[i] / (float) (1<<15); + isf_q[i] += PRED_FACTOR * isf_past[i]; isf_past[i] = tmp; } } commit 50f3a13500de12d7d9fb3b382965c133a9eea200 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Wed Jul 21 14:21:49 2010 -0300 Use separate states for the high-pass filters diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 063afc0..992bcdc 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -74,7 +74,7 @@ typedef struct { float samples_in[LP_ORDER + AMRWB_SUBFRAME_SIZE]; ///< floating point samples float demph_mem[1]; ///< previous value in the de-emphasis filter - float hpf_mem[4]; ///< previous values in the high-pass filter + float hpf_31_mem[4], hpf_400_mem[4]; ///< previous values in the high-pass filters AVLFG prng; ///< random number generator for white noise excitation } AMRWBContext; @@ -1150,14 +1150,14 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, de_emphasis(&ctx->samples_in[LP_ORDER], PREEMPH_FAC, ctx->demph_mem); high_pass_filter(&ctx->samples_in[LP_ORDER], hpf_31_coef, - ctx->hpf_mem, &ctx->samples_in[LP_ORDER]); + ctx->hpf_31_mem, &ctx->samples_in[LP_ORDER]); // XXX: the 5/4 upsampling for the lower band goes in here /* High frequency band generation */ high_pass_filter(&ctx->samples_in[LP_ORDER], hpf_400_coef, - ctx->hpf_mem, &ctx->samples_in[LP_ORDER]); + ctx->hpf_400_mem, &ctx->samples_in[LP_ORDER]); hb_gain = find_hb_gain(ctx, &ctx->samples_in[LP_ORDER], cur_subframe->hb_gain, cf->vad); ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdata.h | 16 +++----- libavcodec/amrwbdec.c | 95 +++++++++++++++++++++++++++++++++++++----------- 2 files changed, 80 insertions(+), 31 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 21 Jul '10

21 Jul '10

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "AMR-WB decoder". The branch, master has been updated via c2a8e2ad2d9c51c9d1f7b9101175db05f2fecabb (commit) from 958f5f8c6f85ce8db4f9b583070804d9a019bd46 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit c2a8e2ad2d9c51c9d1f7b9101175db05f2fecabb Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Wed Jul 21 01:46:19 2010 -0300 Generate the gain-scaled high band excitation (6.3.1) diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 2fcc61a..063afc0 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -19,6 +19,8 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ +#include "libavutil/lfg.h" + #include "avcodec.h" #include "get_bits.h" #include "lsp.h" @@ -73,6 +75,8 @@ typedef struct { float demph_mem[1]; ///< previous value in the de-emphasis filter float hpf_mem[4]; ///< previous values in the high-pass filter + + AVLFG prng; ///< random number generator for white noise excitation } AMRWBContext; static av_cold int amrwb_decode_init(AVCodecContext *avctx) @@ -82,6 +86,8 @@ static av_cold int amrwb_decode_init(AVCodecContext *avctx) avctx->sample_fmt = SAMPLE_FMT_FLT; + av_lfg_init(&ctx->prng, 1); + ctx->excitation = &ctx->excitation_buf[PITCH_MAX + LP_ORDER + 1]; for (i = 0; i < LP_ORDER; i++) { @@ -876,17 +882,17 @@ static void pitch_enhancer(float *fixed_vector, float voice_fac) * * @param ctx [in] pointer to the AMRWBContext * @param lpc [in] pointer to the LPC coefficients + * @param excitation [out] buffer for synthesis final excitation * @param fixed_gain [in] fixed codebook gain for synthesis * @param fixed_vector [in] algebraic codebook vector * @param samples [out] pointer to the output speech samples * @param overflow [in] 16-bit predicted overflow flag */ -static uint8_t synthesis(AMRWBContext *ctx, float *lpc, +static uint8_t synthesis(AMRWBContext *ctx, float *lpc, float *excitation, float fixed_gain, const float *fixed_vector, float *samples, uint8_t overflow) { int i; - float excitation[AMRWB_SUBFRAME_SIZE]; // if an overflow has been detected, the pitch vector is scaled down by a // factor of 4 @@ -998,11 +1004,39 @@ static float find_hb_gain(AMRWBContext *ctx, const float *synth, tilt = ff_dot_productf(synth, synth + 1, AMRWB_SUBFRAME_SIZE - 1) / ff_dot_productf(synth, synth, AMRWB_SUBFRAME_SIZE); + tilt = FFMAX(0.0, tilt); + /* return gain bounded by [0.1, 1.0] */ return FFMAX(0.1, FFMIN(1.0, (1.0 - tilt) * (1.25 - 0.25 * wsp))); } /** + * Generate the high band excitation with the same energy from the lower + * one and scaled by the given gain + * + * @param ctx [in] the context + * @param hb_exc [out] buffer for the excitation + * @param synth_exc [in] excitation used for synthesis + * @param hb_gain [in] wanted excitation gain + */ +static void scaled_hb_excitation(AMRWBContext *ctx, float *hb_exc, + const float *synth_exc, float hb_gain) +{ + int i; + float energy = ff_dot_productf(synth_exc, synth_exc, AMRWB_SUBFRAME_SIZE); + + /* Generate a white-noise excitation */ + for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) + hb_exc[i] = 32768.0 - (uint16_t) av_lfg_get(&ctx->prng) / 65536.0; + + ff_scale_vector_to_given_sum_of_squares(hb_exc, hb_exc, energy, + AMRWB_SUBFRAME_SIZE); + + for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) + hb_exc[i] *= hb_gain; +} + +/** * Update context state before the next subframe */ static void update_sub_state(AMRWBContext *ctx) @@ -1024,6 +1058,8 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, float *synth_fixed_vector; // pointer to the fixed vector that synthesis should use float synth_fixed_gain; // the fixed gain that synthesis should use float voice_fac, stab_fac; // parameters used for gain smoothing + float synth_exc[AMRWB_SUBFRAME_SIZE]; // post-processed excitation for synthesis + float hb_exc[AMRWB_SUBFRAME_SIZE]; // excitation for the high frequency band float hb_gain; int sub, i; @@ -1102,12 +1138,12 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, pitch_enhancer(synth_fixed_vector, voice_fac); - if (synthesis(ctx, ctx->lp_coef[sub], synth_fixed_gain, + if (synthesis(ctx, ctx->lp_coef[sub], synth_exc, synth_fixed_gain, synth_fixed_vector, &ctx->samples_in[LP_ORDER], 0)) // overflow detected -> rerun synthesis scaling pitch vector down // by a factor of 4, skipping pitch vector contribution emphasis // and adaptive gain control - synthesis(ctx, ctx->lp_coef[sub], synth_fixed_gain, + synthesis(ctx, ctx->lp_coef[sub], synth_exc, synth_fixed_gain, synth_fixed_vector, &ctx->samples_in[LP_ORDER], 1); /* Synthesis speech post-processing */ @@ -1126,6 +1162,8 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, hb_gain = find_hb_gain(ctx, &ctx->samples_in[LP_ORDER], cur_subframe->hb_gain, cf->vad); + scaled_hb_excitation(ctx, hb_exc, synth_exc, hb_gain); + /* Update buffers and history */ ff_clear_fixed_vector(ctx->fixed_vector, &fixed_sparse, AMRWB_SUBFRAME_SIZE); ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdec.c | 46 ++++++++++++++++++++++++++++++++++++++++++---- 1 files changed, 42 insertions(+), 4 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 21 Jul '10

21 Jul '10

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "AMR-WB decoder". The branch, master has been updated via 958f5f8c6f85ce8db4f9b583070804d9a019bd46 (commit) via 31a3641e4876dfb07633068be482ed1c140808a6 (commit) from ef5497e0213b4143f745be78fdd2dfc3579dee8c (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit 958f5f8c6f85ce8db4f9b583070804d9a019bd46 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Wed Jul 21 00:15:59 2010 -0300 Begin high band generation by its excitation gain calculation diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 42d5655..29624fa 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -60,7 +60,7 @@ typedef struct { uint16_t adap; ///< adaptive codebook index uint16_t ltp; ///< ltp-filtering flag uint16_t vq_gain; ///< VQ adaptive and innovative gains - uint16_t energy; ///< high-band energy + uint16_t hb_gain; ///< high-band energy index (mode 23k85 only) uint16_t pul_ih[4]; ///< MSBs part of codebook index (high modes only) uint16_t pul_il[4]; ///< LSBs part of codebook index } AMRWBSubFrame; @@ -597,7 +597,7 @@ static const uint16_t order_MODE_23k85[] = { 11, AMR_OF(0, pul_il[3]), 234, 192, 358, 448, 260, 239, 268, 373, 401, 428, 473, 7, AMR_OF(0, vq_gain), 3, 20, 42, 28, 32, 38, 24, - 4, AMR_OF(0, energy), 72, 73, 74, 75, + 4, AMR_OF(0, hb_gain), 72, 73, 74, 75, 6, AMR_OF(1, adap), 36, 49, 88, 93, 99, 114, 1, AMR_OF(1, ltp), 122, 11, AMR_OF(1, pul_ih[0]), 132, 155, 153, 182, 241, 308, 360, 252, @@ -617,7 +617,7 @@ static const uint16_t order_MODE_23k85[] = { 11, AMR_OF(1, pul_il[3]), 246, 215, 368, 451, 269, 277, 309, 402, 423, 435, 471, 7, AMR_OF(1, vq_gain), 4, 21, 43, 29, 33, 39, 25, - 4, AMR_OF(1, energy), 76, 77, 78, 79, + 4, AMR_OF(1, hb_gain), 76, 77, 78, 79, 9, AMR_OF(2, adap), 15, 16, 17, 18, 19, 51, 70, 96, 108, 1, AMR_OF(2, ltp), 123, @@ -638,7 +638,7 @@ static const uint16_t order_MODE_23k85[] = { 11, AMR_OF(2, pul_il[3]), 303, 205, 399, 450, 283, 275, 264, 379, 425, 438, 462, 7, AMR_OF(2, vq_gain), 5, 22, 44, 30, 34, 40, 26, - 4, AMR_OF(2, energy), 80, 81, 82, 83, + 4, AMR_OF(2, hb_gain), 80, 81, 82, 83, 6, AMR_OF(3, adap), 37, 50, 89, 94, 100, 115, 1, AMR_OF(3, ltp), 124, 11, AMR_OF(3, pul_ih[0]), 135, 168, 152, 179, 226, 302, 347, 299, @@ -658,7 +658,7 @@ static const uint16_t order_MODE_23k85[] = { 11, AMR_OF(3, pul_il[3]), 255, 183, 389, 449, 244, 281, 305, 394, 405, 434, 460, 7, AMR_OF(3, vq_gain), 6, 23, 45, 31, 35, 41, 27, - 4, AMR_OF(3, energy), 84, 85, 86, 87, + 4, AMR_OF(3, hb_gain), 84, 85, 86, 87, 0 }; @@ -1792,12 +1792,23 @@ static const float *ir_filters_lookup[2] = { ir_filter_str, ir_filter_mid }; -/* High-pass filter coefficients for inputs and outputs (feedback) */ -static const float hpf_x_coef[3] = { - 0.989501953, -1.979003906, 0.989501953 +/* High-pass filters coefficients for inputs and outputs (feedback) */ + +static const float hpf_31_coef[2][3] = { // 31 kHz cutoff filter + { 0.989501953, -1.979003906, 0.989501953 }, + { 1.978881836, -0.979125977, 0 } +}; + +static const float hpf_400_coef[2][3] = { // 400 kHz cutoff filter + { 0.893554687, -1.787109375, 0.893554687 }, + { 1.787109375, -0.864257812, 0 } }; -static const float hpf_y_coef[2] = { 1.978881836, -0.979125977 }; +/* High band quantized gains for 23k85 in Q14 */ +static const uint16_t qua_hb_gain[16] = { + 3624, 4673, 5597, 6479, 7425, 8378, 9324, 10264, + 11210, 12206, 13391, 14844, 16770, 19655, 24289, 32728 +}; /* Core frame sizes in each mode */ static const uint16_t cf_sizes_wb[] = { diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index dac45d3..2fcc61a 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -950,12 +950,14 @@ static void de_emphasis(float *synth, float m, float mem[1]) * with cutoff frequency at 31 Hz * * @param out [out] buffer for filtered output + * @param hpf_coef [in] filter coefficients as used below * @param mem [in] state from last filtering * @param in [in] input speech data * * @remark It is safe to pass the same array in in and out parameters. */ -static void high_pass_filter(float *out, float mem[4], const float *in) +static void high_pass_filter(float *out, const float hpf_coef[2][3], + float mem[4], const float *in) { int i; float *x = mem - 1, *y = mem + 2; // previous inputs and outputs @@ -963,9 +965,9 @@ static void high_pass_filter(float *out, float mem[4], const float *in) for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) { float x0 = in[i]; - out[i] = hpf_x_coef[0] * x0 + hpf_y_coef[0] * y[0] + - hpf_x_coef[1] * x[1] + hpf_y_coef[1] * y[1] + - hpf_x_coef[2] * x[2]; + out[i] = hpf_coef[0][0] * x0 + hpf_coef[1][0] * y[0] + + hpf_coef[0][1] * x[1] + hpf_coef[1][1] * y[1] + + hpf_coef[0][2] * x[2]; y[1] = y[0]; y[0] = out[i]; @@ -976,6 +978,31 @@ static void high_pass_filter(float *out, float mem[4], const float *in) } /** + * Calculate the high band gain based on encoded index (23k85 mode) or + * on the lower band speech signal and the Voice Activity Detection flag + * + * @param ctx [in] the context + * @param synth [in] LB speech synthesis at 12.8k + * @param hb_idx [in] gain index for mode 23k85 only + * @param vad [in] VAD flag for the frame + */ +static float find_hb_gain(AMRWBContext *ctx, const float *synth, + uint16_t hb_idx, uint8_t vad) +{ + int wsp = (vad > 0 ? 1 : 0); + float tilt; + + if (ctx->fr_cur_mode == MODE_23k85) + return qua_hb_gain[hb_idx] / (float) (1 << 14); + + tilt = ff_dot_productf(synth, synth + 1, AMRWB_SUBFRAME_SIZE - 1) / + ff_dot_productf(synth, synth, AMRWB_SUBFRAME_SIZE); + + /* return gain bounded by [0.1, 1.0] */ + return FFMAX(0.1, FFMIN(1.0, (1.0 - tilt) * (1.25 - 0.25 * wsp))); +} + +/** * Update context state before the next subframe */ static void update_sub_state(AMRWBContext *ctx) @@ -997,6 +1024,7 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, float *synth_fixed_vector; // pointer to the fixed vector that synthesis should use float synth_fixed_gain; // the fixed gain that synthesis should use float voice_fac, stab_fac; // parameters used for gain smoothing + float hb_gain; int sub, i; ctx->fr_cur_mode = unpack_bitstream(ctx, buf, buf_size); @@ -1085,8 +1113,18 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, /* Synthesis speech post-processing */ de_emphasis(&ctx->samples_in[LP_ORDER], PREEMPH_FAC, ctx->demph_mem); - high_pass_filter(&ctx->samples_in[LP_ORDER], ctx->hpf_mem, - &ctx->samples_in[LP_ORDER]); + high_pass_filter(&ctx->samples_in[LP_ORDER], hpf_31_coef, + ctx->hpf_mem, &ctx->samples_in[LP_ORDER]); + + // XXX: the 5/4 upsampling for the lower band goes in here + + + /* High frequency band generation */ + high_pass_filter(&ctx->samples_in[LP_ORDER], hpf_400_coef, + ctx->hpf_mem, &ctx->samples_in[LP_ORDER]); + + hb_gain = find_hb_gain(ctx, &ctx->samples_in[LP_ORDER], + cur_subframe->hb_gain, cf->vad); /* Update buffers and history */ ff_clear_fixed_vector(ctx->fixed_vector, &fixed_sparse, commit 31a3641e4876dfb07633068be482ed1c140808a6 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Tue Jul 20 02:03:21 2010 -0300 Fix the direct variable indexing inside (sub)frame diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 779fa70..42d5655 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -54,17 +54,19 @@ enum Mode { NO_DATA ///< no transmission }; +/* All decoded parameters in these structs must be 2 bytes long + * because of the direct indexing at the frame parsing */ typedef struct { uint16_t adap; ///< adaptive codebook index - uint8_t ltp; ///< ltp-filtering flag - uint8_t vq_gain; ///< VQ adaptive and innovative gains - uint8_t energy; ///< high-band energy + uint16_t ltp; ///< ltp-filtering flag + uint16_t vq_gain; ///< VQ adaptive and innovative gains + uint16_t energy; ///< high-band energy uint16_t pul_ih[4]; ///< MSBs part of codebook index (high modes only) uint16_t pul_il[4]; ///< LSBs part of codebook index } AMRWBSubFrame; typedef struct { - uint8_t vad; ///< voice activity detection flag + uint16_t vad; ///< voice activity detection flag uint16_t isp_id[7]; ///< index of ISP subvectors AMRWBSubFrame subframe[4]; ///< data for subframes } AMRWBFrame; diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index b1f2cc2..dac45d3 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -124,7 +124,7 @@ static enum Mode unpack_bitstream(AMRWBContext *ctx, const uint8_t *buf, // XXX: We are using only the "MIME/storage" format // used by libopencore. This format is simpler and - // do not have the auxiliary information of the frame + // does not have the auxiliary information of the frame /* AMR-WB Auxiliary Information */ /* ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdata.h | 37 +++++++++++++++++++++++----------- libavcodec/amrwbdec.c | 52 +++++++++++++++++++++++++++++++++++++++++------ 2 files changed, 70 insertions(+), 19 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0

[soc]: r5866 - mms/mmst.c
by spyfeng 20 Jul '10

20 Jul '10

Author: spyfeng Date: Tue Jul 20 17:30:36 2010 New Revision: 5866 Log: synchronize mmst.c with ffmpeg trunk, since we submitted the patches to ffmpeg-devel directly. Modified: mms/mmst.c Modified: mms/mmst.c ============================================================================== --- mms/mmst.c Tue Jul 20 17:26:29 2010 (r5865) +++ mms/mmst.c Tue Jul 20 17:30:36 2010 (r5866) @@ -20,6 +20,14 @@ * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ + +/*References + * MMS protocol specification: + * [1]http://msdn.microsoft.com/en-us/library/cc234711(PROT.10).aspx + *asf specification. Revision 01.20.03. + * [2]http://msdn.microsoft.com/en-us/library/bb643323.aspx + */ + #include "avformat.h" #include "internal.h" #include "libavutil/intreadwrite.h" @@ -282,8 +290,7 @@ static MMSSCPacketType get_tcp_server_re return -1; } hr = AV_RL32(mms->in_buffer + 40); - if (hr) - { + if (hr) { dprintf(NULL, "The server side send back error code:0x%x\n", hr); return -1; } @@ -336,6 +343,7 @@ static MMSSCPacketType get_tcp_server_re mms->remaining_in_len); mms->asf_header_size += mms->remaining_in_len; } + // 0x04 means asf header is sent in multiple packets. if (mms->incoming_flags == 0x04) continue; } else if(packet_id_type == mms->packet_id) { @@ -382,6 +390,7 @@ static int mms_safe_send_recv(MMSContext return ret; } } + if ((type = get_tcp_server_response(mms)) != expect_type) { dprintf(NULL,"Unexpected packet type %d with type %d\n", type, expect_type); return -1; @@ -433,8 +442,7 @@ static int asf_header_parser(MMSContext { uint8_t *p = mms->asf_header; uint8_t *end; - int st_idx = 0; - int flags, stream_id, is_stream_num_known = 0; + int flags, stream_id; mms->stream_num = 0; if (mms->asf_header_size < sizeof(ff_asf_guid) * 2 + 22 || @@ -448,7 +456,7 @@ static int asf_header_parser(MMSContext uint64_t chunksize = AV_RL64(p + sizeof(ff_asf_guid)); if (!chunksize || chunksize > end - p) { dprintf(NULL, "chunksize is exceptional value:%"PRId64"!\n", chunksize); - return 0; + return -1; } if (!memcmp(p, ff_asf_file_header, sizeof(ff_asf_guid))) { /* read packet size */ @@ -468,30 +476,14 @@ static int asf_header_parser(MMSContext //Please see function send_stream_selection_request(). if (mms->stream_num < MAX_STREAMS && 46 + mms->stream_num * 6 < sizeof(mms->out_buffer)) { - st_idx = is_stream_num_known ? st_idx++ : mms->stream_num; - mms->streams[st_idx].id = stream_id; - if (!is_stream_num_known) - mms->stream_num++; + mms->streams[mms->stream_num].id = stream_id; + mms->stream_num++; } else { dprintf(NULL, "Too many streams.\n"); return -1; } } else if (!memcmp(p, ff_asf_head1_guid, sizeof(ff_asf_guid))) { - chunksize = 46; - } else if (!memcmp(p, ff_asf_stream_bitrate_properties, sizeof(ff_asf_guid))) { - int record_cnt = AV_RL16(p + sizeof(ff_asf_guid) + 8); - if (record_cnt*6 + 16 + 8 + 2 > chunksize) { - dprintf(NULL, "Too many stream record count.\n"); - return -1; - } - if (mms->stream_num < MAX_STREAMS && - 46 + mms->stream_num * 6 < sizeof(mms->out_buffer)) { - mms->stream_num = record_cnt; - is_stream_num_known = 1; - } else { - dprintf(NULL, "Too many streams(bitrate properties)\n"); - return -1; - } + chunksize = 46; // see references [2] section 3.4. This should be set 46. } p += chunksize; }

1 0

[soc]: r5865 - mms/mmsh.c
by spyfeng 20 Jul '10

20 Jul '10

Author: spyfeng Date: Tue Jul 20 17:26:29 2010 New Revision: 5865 Log: make mmsh.c resuse http uri, maybe need more test link. work with the links: [1]mms://s1.104mfonline.com.hk/Encoder5 [2]mms://194.116.83.15/70s Modified: mms/mmsh.c Modified: mms/mmsh.c ============================================================================== --- mms/mmsh.c Mon Jul 19 01:20:46 2010 (r5864) +++ mms/mmsh.c Tue Jul 20 17:26:29 2010 (r5865) @@ -24,6 +24,7 @@ #include <string.h> #include "libavutil/avstring.h" #include "asf.h" +#include "http.h" #define DEBUG #define CHUNK_TYPE_DATA 0x4424 @@ -38,7 +39,6 @@ #define CLIENTGUID "Pragma: xClientGUID={c77e7400-738a-11d2-9add-0020af0a3278}\r\n" static const char* mmsh_first_request = - "GET %s HTTP/1.0\r\n" "Accept: */*\r\n" USERAGENT "Host: %s:%d\r\n" @@ -47,7 +47,6 @@ static const char* mmsh_first_request = "Connection: Close\r\n\r\n"; static const char* mmsh_seekable_request = - "GET %s HTTP/1.0\r\n" "Accept: */*\r\n" USERAGENT "Host: %s:%d\r\n" @@ -59,7 +58,6 @@ static const char* mmsh_seekable_request "Connection: Close\r\n\r\n"; static const char* mmsh_live_request = - "GET %s HTTP/1.0\r\n" "Accept: */*\r\n" USERAGENT "Host: %s:%d\r\n" @@ -108,97 +106,6 @@ static int mmsh_close(URLContext *h) return 0; } -static int send_pack(MMSHContext *mms) -{ - int len, result; - len = strlen(mms->out_buffer); - result = url_write(mms->mms_hd, mms->out_buffer, len); - if(result != len) { - dprintf(NULL,"send pack failed!return len %d != %d\n", result, len); - return AVERROR_IO; - } - return 0; -} - -static char* find_str(char * dst, const char *str, const int len) -{ - char *p = NULL; - if(strncasecmp(dst, str, len) == 0) { - p = &dst[len]; - while(isspace(*p)) - p++; - } - return p; -} - -static int get_and_parse_http_header(MMSHContext *mms) -{ - int len = 0, line_num = 0; - int http_code; - char content_type[128]={'\0'}; - char *p, *pos; - //uint8_t tmp_buf[8192]; - //url_read_complete(mms->mms_hd, tmp_buf, sizeof(tmp_buf)); - for(;;) { - if(url_read(mms->mms_hd, &mms->in_buffer[len], 1) != 1) { - dprintf(NULL, "recv http header failed!\n"); - return AVERROR(EIO); - } - - if(mms->in_buffer[len] != 0x0A) { - len++; - if(len >= sizeof(mms->in_buffer)) { - dprintf(NULL, "recv http header overwrite the buffer!\n"); - return -1; - } - } else { - mms->in_buffer[len--] = '\0'; - if (len >= 0 && mms->in_buffer[len] == 0x0D) { - line_num++; - mms->in_buffer[len--] = '\0'; - if(len < 0) { - return 0; // \r\n\r\n is the end of http header - } else { - len = 0;// begin to read next http header line - } - } - if (line_num == 1) { - p = mms->in_buffer; - while(!isspace(*p) && *p != '\0') - p++; - while(isspace(*p)) - p++; - http_code = strtol(p, NULL, 10); - dprintf(NULL, "mmsh protocol http_code=%d\n", http_code); - if(http_code != 200) { - return -1; - } - } else { - if ((p = find_str(mms->in_buffer, "Content-Type:", 13)) != NULL) { - strncpy(content_type, p, sizeof(content_type)); - dprintf(NULL, "Content-Type:%s\n", content_type); - if(strcmp(content_type, "application/x-mms-framed") != 0 - && strcmp(content_type, "application/octet-stream") != 0 - && strcmp(content_type, "application/vnd.ms.wms-hdr.asfv1") != 0) { - return -1; - } - } else if((p = find_str(mms->in_buffer, "Pragma:", 7)) != NULL) { - pos = strstr(p, "features="); - if (pos){ - if(strstr(pos, "seekable")) { - mms->seekable = 1; - } else if (strstr(pos, "broadcast")) { - mms->seekable = 0; - } - } else { - dprintf(NULL, "Can't find features!\n"); - } - } - } - } - } -} - static int asf_header_parser(MMSHContext * mms) { uint8_t *p = mms->asf_header; @@ -217,7 +124,7 @@ static int asf_header_parser(MMSHContext while(end - p >= sizeof(ff_asf_guid) + 8) { uint64_t chunksize = AV_RL64(p + sizeof(ff_asf_guid)); if (!chunksize || chunksize > end - p) { - dprintf(NULL, "chunksize is exceptional value:%d!\n", chunksize); + dprintf(NULL, "chunksize is exceptional value:%"PRId64"!\n", chunksize); return -1; } if (!memcmp(p, ff_asf_file_header, sizeof(ff_asf_guid))) { @@ -359,53 +266,46 @@ static int get_http_header_data(MMSHCont return 0; } -static int get_http_answer(MMSHContext *mms) -{ - int result; - result = get_and_parse_http_header(mms); - if (result) { - dprintf(NULL, "http header parser failed!\n"); - return result; - } - - result = get_http_header_data(mms); - if (result) { - dprintf(NULL, "get http header data fialed!\n"); - return result; - } - return 0; -} - static int mmsh_open_cnx(MMSHContext *mms) { int i, port, err, offset = 0; - char tcpname[256], path[256], host[128]; + char httpname[256], path[256], host[128]; char stream_selection[10 * MAX_STREAMS]; + char headers[1024]; if (mms->mms_hd) { url_close(mms->mms_hd); } + ff_url_split(NULL, 0, NULL, 0, host, sizeof(host), &port, path, sizeof(path), mms->location); if(port<0) port = 80; // defaut mmsh protocol port + ff_url_join(httpname, sizeof(httpname), "http", NULL, host, port, path); + + if (url_alloc(&mms->mms_hd, httpname, URL_RDONLY) < 0) { + return AVERROR(EIO); + } + + snprintf (headers, sizeof(headers), mmsh_first_request, + host, port, mms->request_seq++); + ff_http_set_headers(mms->mms_hd, headers); + + if (url_connect(mms->mms_hd)) { + return AVERROR(EIO); + } + err = get_http_header_data(mms); + if (err) { + dprintf(NULL, "get http header data fialed!\n"); + return (err); + } - ff_url_join(tcpname, sizeof(tcpname), "tcp", NULL, host, port, NULL); - err = url_open(&mms->mms_hd, tcpname, URL_RDWR); - if (err) - return err; - // send describe request - snprintf (mms->out_buffer, sizeof(mms->out_buffer), mmsh_first_request, path, - host, port, mms->request_seq++); - err = send_pack(mms); - if (err) - return err; - err = get_http_answer(mms); // TODO match with the first request - if(err) - return err; // close the socket and then reopen it for sending the second play request. url_close(mms->mms_hd); - + memset(headers, 0, sizeof(headers)); + if (url_alloc(&mms->mms_hd, httpname, URL_RDONLY) < 0) { + return AVERROR(EIO); + } for (i = 0; i < mms->stream_num; i++) { err = snprintf(stream_selection + offset, sizeof(stream_selection) - offset, "ffff:%d:0 ", mms->streams[i].id); @@ -413,31 +313,31 @@ static int mmsh_open_cnx(MMSHContext *mm return err; offset += err; } - // send paly request if (mms->seekable) { - err = snprintf(mms->out_buffer, sizeof(mms->out_buffer), mmsh_seekable_request, path, + err = snprintf(headers, sizeof(headers), mmsh_seekable_request, host, port, 0, 0, 0, mms->request_seq++, 0, mms->stream_num, stream_selection); } else { - err = snprintf(mms->out_buffer, sizeof(mms->out_buffer), mmsh_live_request, path, + err = snprintf(headers, sizeof(headers), mmsh_live_request, host, port, mms->request_seq++, mms->stream_num, stream_selection); } if (err < 0) { dprintf(NULL, "build play request failed!\n"); return err; } - dprintf(NULL, "out_buffer is %s", mms->out_buffer); + dprintf(NULL, "out_buffer is %s", headers); + ff_http_set_headers(mms->mms_hd, headers); + + if (url_connect(mms->mms_hd)) { + return AVERROR(EIO); + } + + err = get_http_header_data(mms); + if (err) { + dprintf(NULL, "get http header data fialed!\n"); + return (err); + } - //reopen the connection. - err = url_open(&mms->mms_hd, tcpname, URL_RDWR); - if (err) - return err; - err = send_pack(mms); - if (err) - return err; - err = get_http_answer(mms);// TODO mathc with the second request - if(err) - return err; return 0; }

1 0

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 19 Jul '10

19 Jul '10

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "AMR-WB decoder". The branch, master has been updated via ef5497e0213b4143f745be78fdd2dfc3579dee8c (commit) via b6c07e986a96e8b699c79fa69939ff1bdf3f8def (commit) from 4df069267c507ece9555404b369ea6043e116854 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit ef5497e0213b4143f745be78fdd2dfc3579dee8c Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Mon Jul 19 23:16:17 2010 -0300 Fix AVCodec interfacing, now it returns the number of bytes consumed and samples produced diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 21b0a5e..779fa70 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -30,7 +30,8 @@ #define MIN_ENERGY -14.0 ///< initial innnovation energy (dB) #define ENERGY_MEAN 30.0 ///< mean innovation energy (dB) in all modes -#define AMRWB_SUBFRAME_SIZE 64 ///< samples per subframe +#define AMRWB_SUBFRAME_SIZE 64 ///< samples per subframe at 12.8 kHz +#define AMRWB_SFR_SIZE_OUT 80 ///< samples per subframe at 16 kHz #define AMRWB_SAMPLE_BOUND 32768.0 ///< threshold for synthesis overflow #define PITCH_MAX 231 ///< maximum received pitch delay value diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index e9866d8..b1f2cc2 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -75,11 +75,13 @@ typedef struct { float hpf_mem[4]; ///< previous values in the high-pass filter } AMRWBContext; -static int amrwb_decode_init(AVCodecContext *avctx) +static av_cold int amrwb_decode_init(AVCodecContext *avctx) { AMRWBContext *ctx = avctx->priv_data; int i; + avctx->sample_fmt = SAMPLE_FMT_FLT; + ctx->excitation = &ctx->excitation_buf[PITCH_MAX + LP_ORDER + 1]; for (i = 0; i < LP_ORDER; i++) { @@ -1095,7 +1097,10 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, // update state for next frame memcpy(ctx->isp_sub4_past, ctx->isp[3], LP_ORDER * sizeof(ctx->isp[3][0])); - return 0; + /* report how many samples we got */ + *data_size = 4 * AMRWB_SFR_SIZE_OUT * sizeof(float); + + return ((cf_sizes_wb[ctx->fr_cur_mode] + 7) >> 3) + 1; } AVCodec amrwb_decoder = commit b6c07e986a96e8b699c79fa69939ff1bdf3f8def Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Mon Jul 19 23:12:11 2010 -0300 Fix frame parsing, the reordering tables seems to work at least for the isp indexes diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index d2612f4..21b0a5e 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -69,7 +69,7 @@ typedef struct { } AMRWBFrame; /** The index of a frame parameter */ -#define AMR_BIT(field) (offsetof(AMRWBFrame, field)) +#define AMR_BIT(field) (offsetof(AMRWBFrame, field) >> 1) /** The index of a subframe-specific parameter */ #define AMR_OF(frame_num, variable) AMR_BIT(subframe[frame_num].variable) diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 97c806e..e9866d8 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -113,18 +113,23 @@ static enum Mode unpack_bitstream(AMRWBContext *ctx, const uint8_t *buf, init_get_bits(&gb, buf, buf_size * 8); - /* AMR-WB header */ + /* AMR-WB header (1st octet) */ + skip_bits(&gb, 1); // padding bit ctx->fr_cur_mode = get_bits(&gb, 4); mode = ctx->fr_cur_mode; ctx->fr_quality = get_bits1(&gb); + skip_bits(&gb, 2); - skip_bits(&gb, 3); + // XXX: We are using only the "MIME/storage" format + // used by libopencore. This format is simpler and + // do not have the auxiliary information of the frame /* AMR-WB Auxiliary Information */ - ctx->fr_mode_ind = get_bits(&gb, 4); - ctx->fr_mode_req = get_bits(&gb, 4); - // XXX: Need to check conformity in mode_ind/mode_req and crc? - ctx->fr_crc = get_bits(&gb, 8); + /* + * ctx->fr_mode_ind = get_bits(&gb, 4); + * ctx->fr_mode_req = get_bits(&gb, 4); + * ctx->fr_crc = get_bits(&gb, 8); + */ data = (uint16_t *) &ctx->frame; memset(data, 0, sizeof(AMRWBFrame)); @@ -138,9 +143,10 @@ static enum Mode unpack_bitstream(AMRWBContext *ctx, const uint8_t *buf, int field = 0; int field_offset = *perm++; while (field_size--) { - uint16_t bit = *perm++; + uint16_t bit_idx = *perm++; field <<= 1; - field |= buf[bit >> 3] >> (bit & 7) & 1; + /* The bit index inside the byte is reversed */ + field |= BIT_POS(buf[bit_idx >> 3], 7 - (bit_idx & 7)); } data[field_offset] = field; } ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdata.h | 5 +++-- libavcodec/amrwbdec.c | 31 +++++++++++++++++++++---------- 2 files changed, 24 insertions(+), 12 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0

[PATCH] avoid extra memcpy in Xiph RTP depacketizer
by Josh Allmann 19 Jul '10

19 Jul '10

Hi guys, $subj Josh

3 3

[soc]: r5864 - seek2010/seek2010_07_19_r24319.patch
by mchinen 18 Jul '10

18 Jul '10

Author: mchinen Date: Mon Jul 19 01:20:46 2010 New Revision: 5864 Log: Update av_build_index based seeking to work with FLAC. Fix initial table seek to check for invalid start_time and data_offset before seeking. fix broken flag saving in av_save_index add AVSEEK_FLAG_ANY to initial seek in av_build_index Added: seek2010/seek2010_07_19_r24319.patch Added: seek2010/seek2010_07_19_r24319.patch ============================================================================== --- /dev/null 00:00:00 1970 (empty, because file is newly added) +++ seek2010/seek2010_07_19_r24319.patch Mon Jul 19 01:20:46 2010 (r5864) @@ -0,0 +1,565 @@ +Index: ffplay.c +=================================================================== +--- ffplay.c (revision 24319) ++++ ffplay.c (working copy) +@@ -2381,6 +2381,7 @@ + AVFormatParameters params, *ap = &params; + int eof=0; + int pkt_in_play_range = 0; ++ char *index_filename = NULL; + + ic = avformat_alloc_context(); + +@@ -2497,6 +2498,18 @@ + goto fail; + } + ++ if(ic->filename){ ++ index_filename = av_malloc(strlen(ic->filename) + 5); ++ strcpy(index_filename, ic->filename); ++ strcat(index_filename, ".fdx"); ++ } ++ ++ if (av_load_index_file(ic, index_filename) < 0 ) { ++ av_build_index(ic, 0); //AV_BUILD_INDEX_PARALLEL); ++ av_save_index_file(ic, index_filename); ++ } ++ av_free(index_filename); ++ + for(;;) { + if (is->abort_request) + break; +Index: libavformat/avformat.h +=================================================================== +--- libavformat/avformat.h (revision 24319) ++++ libavformat/avformat.h (working copy) +@@ -536,6 +536,16 @@ + * Number of frames that have been demuxed during av_find_stream_info() + */ + int codec_info_nb_frames; ++ ++ /* new av_seek_frame_table() support */ ++#define AV_SEEKTABLE_BUILDING 0x0001 ///< a flag set by av_build_index to mark that the index is being built ++#define AV_SEEKTABLE_CBR 0x0002 ///< a flag set by av_build_index to note that the file is constant bit rate ++#define AV_SEEKTABLE_FINISHED 0x0004 ///< a flag set by av_build_index to note that the complete index table is ready to use ++#define AV_SEEKTABLE_COPIED 0x0008 ///< a flag to note that the seek table has been copied. ++ int seek_table_flags; ++ AVIndexEntry* parallel_index_entries; ///used with AV_BUILD_INDEX_PARALLEL. should not be touched by client ++ int parallel_nb_index_entries; ++ unsigned int parallel_index_entries_allocated_size; + } AVStream; + + #define AV_PROGRAM_RUNNING 1 +@@ -1134,6 +1144,26 @@ + int av_index_search_timestamp(AVStream *st, int64_t timestamp, int flags); + + /** ++ * Builds a complete index for seeking in each stream where it is possible. ++ * Requires that the streams have been opened. ++ * Part of the new seeking api. incomplete. ++ */ ++int av_build_index(AVFormatContext *s, int flags); ++#define AV_BUILD_INDEX_PARALLEL 0x0001 ///< Builds the index via a copied demuxer streams. av_build_index with this flag can be called on a seperate thread while decoding is happening on another. ++ ++/** ++ * Saves the index table built with av_build_index to a file ++ * Returns 0 if succesful or a negative value on failure. ++ */ ++int av_save_index_file(AVFormatContext *ic, const char *filename); ++ ++/** ++ * Loads the index saved with av_save_index. ++ * Returns 0 if successful or a negative value on failure. ++ */ ++int av_load_index_file(AVFormatContext *ic, const char *filename); ++ ++/** + * Ensure the index uses less memory than the maximum specified in + * AVFormatContext.max_index_size by discarding entries if it grows + * too large. +@@ -1152,6 +1182,15 @@ + int size, int distance, int flags); + + /** ++ * Perform a dictionary search on the seek table. ++ * av_build_index must be successfully called before using this function. ++ * @param timestamp target timestamp in the time base of the given stream ++ * @param stream_index stream number ++ */ ++int av_seek_frame_table(AVFormatContext *s, ++ int stream_index, int64_t timestamp, int flags); ++ ++/** + * Perform a binary search using av_index_search_timestamp() and + * AVCodec.read_timestamp(). + * This is not supposed to be called directly by a user application, +Index: libavformat/utils.c +=================================================================== +--- libavformat/utils.c (revision 24319) ++++ libavformat/utils.c (working copy) +@@ -1031,7 +1031,6 @@ + pkt->convergence_duration = pc->convergence_duration; + } + +- + static int av_read_frame_internal(AVFormatContext *s, AVPacket *pkt) + { + AVStream *st; +@@ -1355,21 +1354,28 @@ + + st->index_entries= entries; + +- index= av_index_search_timestamp(st, timestamp, AVSEEK_FLAG_ANY); ++ //if we are building the table, the indicies added in order, so we don't have to do expensive searching. ++ if(st->seek_table_flags & AV_SEEKTABLE_BUILDING) { ++ index = st->nb_index_entries++; ++ ie = &st->index_entries[index]; ++ } ++ else { ++ index= av_index_search_timestamp(st, timestamp, AVSEEK_FLAG_ANY); + +- if(index<0){ +- index= st->nb_index_entries++; +- ie= &entries[index]; +- assert(index==0 || ie[-1].timestamp < timestamp); +- }else{ +- ie= &entries[index]; +- if(ie->timestamp != timestamp){ +- if(ie->timestamp <= timestamp) +- return -1; +- memmove(entries + index + 1, entries + index, sizeof(AVIndexEntry)*(st->nb_index_entries - index)); +- st->nb_index_entries++; +- }else if(ie->pos == pos && distance < ie->min_distance) //do not reduce the distance +- distance= ie->min_distance; ++ if(index<0){ ++ index= st->nb_index_entries++; ++ ie= &entries[index]; ++ assert(index==0 || ie[-1].timestamp < timestamp); ++ }else{ ++ ie= &entries[index]; ++ if(ie->timestamp != timestamp){ ++ if(ie->timestamp <= timestamp) ++ return -1; ++ memmove(entries + index + 1, entries + index, sizeof(AVIndexEntry)*(st->nb_index_entries - index)); ++ st->nb_index_entries++; ++ }else if(ie->pos == pos && distance < ie->min_distance) //do not reduce the distance ++ distance= ie->min_distance; ++ } + } + + ie->pos = pos; +@@ -1705,6 +1711,11 @@ + timestamp = av_rescale(timestamp, st->time_base.den, AV_TIME_BASE * (int64_t)st->time_base.num); + } + ++ /* if we've built a seek table, use it. */ ++ st = s->streams[stream_index]; ++ if (st->seek_table_flags & AV_SEEKTABLE_FINISHED) ++ return av_seek_frame_table(s, stream_index, timestamp, flags); ++ + /* first, we try the format specific seek */ + if (s->iformat->read_seek) + ret = s->iformat->read_seek(s, stream_index, timestamp, flags); +@@ -1742,6 +1753,402 @@ + // try some generic seek like av_seek_frame_generic() but with new ts semantics + } + ++int av_seek_frame_table(AVFormatContext *s, ++ int stream_index, int64_t timestamp, int flags) ++{ ++ int index; ++ int64_t ret; ++ AVStream *st; ++ AVIndexEntry *ie; ++ ++ st = s->streams[stream_index]; ++ ++ //TODO: see if we can do something with the CBR field. ++ if(st->seek_table_flags & AV_SEEKTABLE_CBR){ ++ int64_t target_pos; ++ int64_t start_time; ++ //if there is a constant frame length, we can compute where to jump to, but CBR may not have constant frame length ++ ++ start_time = st->start_time == AV_NOPTS_VALUE ? 0 : st->start_time; ++ target_pos = s->data_offset + s->file_size * (timestamp - start_time) * av_q2d(st->time_base) / (st->duration * av_q2d(st->time_base)); ++ if ((ret = url_fseek(s->pb, target_pos, SEEK_SET)) < 0) ++ return ret; ++ av_update_cur_dts(s, st, timestamp); ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: cbr seek ok to %lli of %lli bytes\n", target_pos, s->file_size); ++ return 0; ++ } ++ ++ //see if we need to move the parallel table over ++ if(st->parallel_index_entries) { ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: copying over parallel frames\n"); ++ av_free(st->index_entries); ++ st->index_entries = st->parallel_index_entries; ++ st->nb_index_entries = st->parallel_nb_index_entries; ++ st->index_entries_allocated_size = st->parallel_index_entries_allocated_size; ++ st->parallel_index_entries = NULL; ++ } ++ ++ index = av_index_search_timestamp(st, timestamp, flags); ++ ++ /* this function should only be called after the table is complete. */ ++ if(index < 0 || index >= st->nb_index_entries-1){ ++ return -1; ++ } ++ ++ ff_read_frame_flush(s); ++ /* we use the native seek function if it exists, (still have to modify them to use seek_table) */ ++ if (s->iformat->read_seek){ ++ if(s->iformat->read_seek(s, stream_index, timestamp, flags) >= 0) { ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: table seeked using native function\n"); ++ return 0; ++ } ++ } ++ ++ ie = &st->index_entries[index]; ++ if ((ret = url_fseek(s->pb, ie->pos, SEEK_SET)) < 0) ++ return ret; ++ av_update_cur_dts(s, st, ie->timestamp); ++ ++ { ++ float request_time = (((float)timestamp/st->time_base.den)*st->time_base.num); ++ float actual_time = (((float)ie->timestamp/st->time_base.den)*st->time_base.num); ++ float time_base_inv = ((float)st->time_base.den/st->time_base.num); ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: table seeked to %.2fs (actual request was %.2fs), timebaseinv %f, samplerate of %d\n",actual_time, request_time, time_base_inv, st->codec->sample_rate ); ++ } ++ return 0; ++} ++ ++//if we don't have the number of frames, we just guess. ++#define AV_DETECT_CBR_FPS_GUESS 10 ++static int av_detect_cbr(AVFormatContext *s, AVStream *st) { ++ //check the stream's codec to see if a number of frames are listed, but we can't assume that is available. ++ if(s->data_offset > 0 && s->file_size > 0 && s->data_offset != s->file_size && st->codec && st->duration > 0) { ++ double sec_duration = st->duration * av_q2d(st->time_base); ++ int nb_frames = st->nb_frames?st->nb_frames:(sec_duration*AV_DETECT_CBR_FPS_GUESS); ++ int64_t br_dur_len = (st->codec->bit_rate * sec_duration)/8; ++ int64_t min_len = (s->file_size - s->data_offset) * (1 - 2.0/nb_frames); ++ int64_t max_len = (s->file_size - s->data_offset) * (1 + 2.0/nb_frames); ++ ++ if(min_len < br_dur_len && br_dur_len < max_len) { ++ av_log(s, AV_LOG_DEBUG, "cbr ok (%lli < %lli < %lli) with bitrate=%i, timebase=%i/%i, duration = %f(%lli), nb_frames = %lli, offset= %lli\n", ++ min_len, br_dur_len, max_len, st->codec->bit_rate,st->time_base.num,st->time_base.den, st->duration*av_q2d(st->time_base),st->duration,st->nb_frames,s->data_offset); ++ return 1; ++ } ++ av_log(s, AV_LOG_DEBUG, "cbr failed !(%lli < %lli < %lli) with bitrate=%i, timebase=%i/%i, duration = %f(%lli), nb_frames = %lli, offset= %lli\n", ++ min_len, br_dur_len, max_len, st->codec->bit_rate,st->time_base.num,st->time_base.den, st->duration*av_q2d(st->time_base),st->duration,st->nb_frames,s->data_offset); ++ }else { ++ av_log(s, AV_LOG_DEBUG, "no cbr test with bitrate=%i, timebase=%i/%i, duration = %f(%lli), nb_frames = %lli, offset= %lli\n", ++ st->codec?st->codec->bit_rate:-1,st->time_base.num,st->time_base.den, st->duration*av_q2d(st->time_base),st->duration,st->nb_frames,s->data_offset); ++ } ++ ++ return 0; ++} ++ ++/** ++ * Starts building a index for seeking. ++ * TODO: use a different file pointer so we can do this in a thread-safe manner, ++ * as there are cases when the user will want to playback while building the index ++ **/ ++int av_build_index(AVFormatContext *s, int flags) ++{ ++ AVStream *st; ++ int ret; ++ int stream_index; ++ int i; ++ int orig_flags; ++ ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: starting building index\n"); ++ stream_index = av_find_default_stream_index(s); ++ if(stream_index < 0) ++ return -1; ++ ++ st = s->streams[stream_index]; ++ /* TODO: check if this stream is CBR and the codec has a seek by timestamp. */ ++ /* if this is true then we can set a flag and exit here. */ ++ if(0 && av_detect_cbr(s,st)) { ++ st->seek_table_flags |= AV_SEEKTABLE_CBR | AV_SEEKTABLE_FINISHED; ++ } else if(0) { ++ /* TODO: for this case see if we have a special method for generating the table */ ++ /* specific to a given format. */ ++ } else if(st->nb_frames!=0 && st->nb_index_entries > st->nb_frames/2) { ++ /* some demuxers load a complete index upon file open. */ ++ st->seek_table_flags |= AV_SEEKTABLE_COPIED; ++ } else { ++ AVFormatContext *build_ic; ++ AVPacket pkt; ++ ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: building index from scratch\n"); ++ ++ /* if the client needs it to be threadsafe, create a new format context to read from. */ ++ if(flags & AV_BUILD_INDEX_PARALLEL) { ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: making thread-safe copy of streams\n"); ++ build_ic = avformat_alloc_context(); ++ ret = av_open_input_file(&build_ic, s->filename, s->iformat, 0, NULL); ++ ++ if(ret < 0) { ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: error re-opening file/streams: %i\n", ret); ++ goto cleanup; ++ } ++ if(build_ic->nb_streams != s->nb_streams) { ++ ret = -1; ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: cloned AVFormatContext has different number of streams!"); ++ goto cleanup; ++ } ++ ++ for(i = 0; i < build_ic->nb_streams; i++) { ++ AVStream *build_st= build_ic->streams[i]; ++ AVCodecContext *avctx = build_st->codec; ++ AVCodec *pcodec; ++ build_ic->streams[i]->discard = AVDISCARD_DEFAULT; ++ ++ //compare with the orignal stream's context, and if opened, copy settings and open the clone ++ if(s->streams[i]->codec->priv_data) { ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: copying stream based on priv_data\n"); ++ if((ret = avcodec_copy_context(avctx, s->streams[i]->codec)) < 0) { ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: error copying codec:%i\n", ret); ++ goto cleanup; ++ } ++ pcodec = avcodec_find_decoder(avctx->codec_id); ++ if((ret = avcodec_open(avctx,pcodec)) < 0) { ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: error opening codec:%i\n", ret); ++ goto cleanup; ++ } ++ } ++ } ++ } else { ++ build_ic = s; ++ } ++ ++ /* default table generation behavior from av_seek_frame_generic */ ++ /* TODO: see why s->data_offset is the file length for avi/mp4 and others */ ++ if ((ret = url_fseek(build_ic->pb, 0/*s->data_offset*/, SEEK_SET)) < 0){ ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: error building index: %i\n", ret); ++ goto cleanup; ++ } ++ ++ orig_flags = build_ic->flags; ++ build_ic->flags |= AVFMT_FLAG_GENPTS; ++ for(i=0;; i++) { ++ do{ ++ ret = av_read_frame(build_ic, &pkt); ++ }while(ret == AVERROR(EAGAIN)); ++ if(ret<0) ++ break; ++ av_free_packet(&pkt); ++ } ++ build_ic->flags = orig_flags; ++ ret = 0; ++ cleanup: ++ if(flags & AV_BUILD_INDEX_PARALLEL) { ++ //TODO: delay/wait till the main index signals us that we are okay to swap ++ if(build_ic) { ++ //take the index over from our clone ++ for(i = 0; i < build_ic->nb_streams; i++) { ++ if(ret >= 0) { ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: marking %i frames from parallel stream ready for copy\n", build_ic->streams[i]->nb_index_entries); ++ s->streams[i]->parallel_index_entries = build_ic->streams[i]->index_entries; ++ s->streams[i]->parallel_nb_index_entries = build_ic->streams[i]->nb_index_entries; ++ s->streams[i]->parallel_index_entries_allocated_size = build_ic->streams[i]->index_entries_allocated_size; ++ build_ic->streams[i]->index_entries = NULL; ++ s->streams[i]->seek_table_flags |= AV_SEEKTABLE_FINISHED; ++ } ++ avcodec_close(build_ic->streams[i]->codec); ++ } ++ av_close_input_file(build_ic); ++ } ++ } ++ if(ret < 0) ++ return ret; ++ } ++ ++ /* since we may have moved the read cursor to the end, return seek to start of stream for non-parallel clients. Not sure if this the desired behavior. */ ++ if( !(flags & AV_BUILD_INDEX_PARALLEL) && !(st->seek_table_flags & AV_SEEKTABLE_COPIED) ) { ++ ff_read_frame_flush(s); ++ for(i=0; i<s->nb_streams;i++) ++ if(s->streams[i]->nb_index_entries) ++ s->streams[i]->seek_table_flags |= AV_SEEKTABLE_FINISHED; ++ ++ if(st->start_time == AV_NOPTS_VALUE || (ret = av_seek_frame(s, stream_index, st->start_time, AVSEEK_FLAG_ANY) ) < 0 ) { ++ int data_offset = (s->data_offset < 0) || (s->data_offset >= s->file_size) ? 0 : s->data_offset; ++ ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: finished building index but error seeking: %i,trying url_fseek\n", ret); ++ ++ /* last ditch effort to seek using the file pointer. */ ++ if ((ret = url_fseek(s->pb, data_offset, SEEK_SET)) < 0) { ++ av_log(s,AV_LOG_DEBUG,"SEEK_TABLE_DEBUG: error seeking with url_fseek: %i\n", ret); ++ return ret; ++ } ++ } ++ } ++ av_log(s, AV_LOG_DEBUG, "SEEK_TABLE_DEBUG: finished building index\n"); ++ return 0; ++} ++ ++/** ++ * Save the flags, size, and differences for the index table ++ */ ++#define AV_SAVEINDEX_PACKETSIZE 1000 ++static int av_save_index_stream(ByteIOContext *bc, AVStream *st) { ++ AVIndexEntry start_ie; ++ AVIndexEntry *last_ie, *curr_ie; ++ int i; ++ ++ //we may need to save something like codec id to be safe. ++ ff_put_v(bc, st->seek_table_flags); ++ //we only write the index if we have a finished table. ++ if(st->seek_table_flags & AV_SEEKTABLE_FINISHED) { ++ ff_put_v(bc, st->nb_index_entries); ++ ++ memset(&start_ie, 0, sizeof(AVIndexEntry)); ++ last_ie = &start_ie; ++ for(i = 0; i < st->nb_index_entries; i++) { ++ curr_ie = &st->index_entries[i]; ++ ff_put_v(bc, curr_ie->pos - last_ie->pos); ++ ff_put_v(bc, curr_ie->timestamp - last_ie->timestamp); ++ ff_put_v(bc, curr_ie->flags - last_ie->flags); ++ ff_put_v(bc, curr_ie->size - last_ie->size); ++ ff_put_v(bc, curr_ie->min_distance - last_ie->min_distance); ++ last_ie = curr_ie; ++ if(i % AV_SAVEINDEX_PACKETSIZE == 0) ++ put_flush_packet(bc); ++ } ++ av_log(NULL, AV_LOG_DEBUG, "successfully wrote index stream with %i entries.\n", st->nb_index_entries); ++ } ++ put_flush_packet(bc); ++ ++ return 0; ++} ++ ++/** ++ * load the flags, size, and differences for the index table ++ */ ++static int av_load_index_stream(ByteIOContext *bc, AVStream *st) { ++ AVIndexEntry *ie, *last_ie; ++ uint64_t v; ++ ++ //check to see if we have a table saved in this stream. ++ v = ff_get_v(bc); ++ if(v & AV_SEEKTABLE_FINISHED) { ++ st->seek_table_flags = v; ++ st->nb_index_entries = v = ff_get_v(bc); ++ ++ //dump the old index and make the new one. ++ av_free(st->index_entries); ++ st->index_entries = av_malloc(v * sizeof(AVIndexEntry)); ++ ++ //the first index entry is not a diff, so it gets special treatment. ++ ie = st->index_entries; ++ if(v--) { ++ ie->pos = ff_get_v(bc); ++ ie->timestamp = ff_get_v(bc); ++ ie->flags = ff_get_v(bc); ++ ie->size = ff_get_v(bc); ++ ie->min_distance = ff_get_v(bc); ++ last_ie = ie; ++ ie++; ++ } ++ ++ while(v--) { ++ ie->pos = ff_get_v(bc) + last_ie->pos; ++ ie->timestamp = ff_get_v(bc) + last_ie->timestamp; ++ ie->flags = ff_get_v(bc) + last_ie->flags; ++ ie->size = ff_get_v(bc) + last_ie->size; ++ ie->min_distance = ff_get_v(bc) + last_ie->min_distance; ++ last_ie = ie; ++ ie++; ++ } ++ av_log(NULL, AV_LOG_DEBUG, "successfully loaded index stream with %i entries.\n", st->nb_index_entries); ++ } ++ return 0; ++} ++ ++/** ++ * Saves the table index built with av_build_index to a file. ++ * Returns 0 if successful, or a negative number if not. ++ * this is not mean to be human-readable ++ * we save the following things in order: ++ * - A strz "FFmpegTableIndex" (17 bytes) ++ * - The libavformat version number ++ * - The file name (null terminated) ++ * - the number of streams (4 bytes) ++ * - each of the streams' index tables via av_save_index_stream ++ */ ++#define AV_INDEX_IDENTIFIER "FFmpegTableIndex" ++#define AV_INDEX_NOFILENAME "nofilename" ++static int av_save_index(ByteIOContext *bc, AVFormatContext *ic) { ++ int i, ret; ++ ++ put_strz(bc, AV_INDEX_IDENTIFIER); ++ ff_put_v(bc, LIBAVFORMAT_VERSION_INT); ++ put_strz(bc, ic->filename?ic->filename:AV_INDEX_NOFILENAME); ++ ff_put_v(bc, ic->nb_streams); ++ for(i = 0; i< ic->nb_streams; i++) { ++ ret = av_save_index_stream(bc, ic->streams[i]); ++ if(ret < 0) ++ return ret; ++ } ++ put_flush_packet(bc); ++ av_log(ic, AV_LOG_DEBUG, "successfully wrote index.\n"); ++ return 0; ++} ++ ++static int av_load_index(ByteIOContext *bc, AVFormatContext *ic) { ++ int i, ret; ++ uint64_t v; ++ char read_str[256]; ++ ++ get_strz(bc, read_str, 255); ++ if( strcmp(read_str, AV_INDEX_IDENTIFIER) ) ++ return -1; ++ //version ++ v = ff_get_v(bc); ++ //make sure the filename matches the context we will write to. ++ get_strz(bc, read_str, 255); ++ if( strlen(read_str) < 255 && strcmp(read_str, ic->filename) && strcmp(read_str, AV_INDEX_NOFILENAME) ) ++ return -1; ++ ++ v = ff_get_v(bc); ++ if( v != ic->nb_streams) ++ return -1; ++ for(i =0; i<v; i++) { ++ ret = av_load_index_stream(bc, ic->streams[i]); ++ if(ret < 0) ++ return ret; ++ } ++ av_log(ic, AV_LOG_DEBUG, "successfully read index.\n"); ++ return 0; ++} ++ ++int av_save_index_file(AVFormatContext *ic, const char* filename) { ++ ByteIOContext* bc; ++ int i, ret, table_completed = 0; ++ ++ //do a first pass to see if we have any tables that need saving. ++ //only write the file if it exists. ++ for(i = 0; i< ic->nb_streams; i++) ++ table_completed |= ic->streams[i]->seek_table_flags & AV_SEEKTABLE_FINISHED; ++ if(!table_completed) ++ return -1; ++ ++ if ((ret=url_fopen(&bc, filename, URL_WRONLY)) < 0) { ++ return -1; ++ } ++ ret = av_save_index(bc, ic); ++ url_fclose(bc); ++ return ret; ++} ++ ++int av_load_index_file(AVFormatContext *ic, const char* filename) { ++ ByteIOContext* bc; ++ int ret; ++ ++ if ((ret=url_fopen(&bc, filename, URL_RDONLY)) < 0) { ++ return -1; ++ } ++ ret = av_load_index(bc, ic); ++ url_fclose(bc); ++ return ret; ++} + /*******************************************************/ + + /**

1 0

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 18 Jul '10

18 Jul '10

This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "AMR-WB decoder". The branch, master has been updated via 4df069267c507ece9555404b369ea6043e116854 (commit) via 89dcf67e516dc9b0b04f890ab8a22b136f6163f7 (commit) via 3c771311f0281d2858c9f53636011069b6e202db (commit) from 87a7325cedff834479d827c0d4b887f1374c36b5 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit 4df069267c507ece9555404b369ea6043e116854 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Sun Jul 18 00:56:40 2010 -0300 Cosmetics: remove all trailing spaces from code diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 8f1d027..d2612f4 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -18,7 +18,7 @@ * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ - + #ifndef AVCODEC_AMRWBDATA_H #define AVCODEC_AMRWBDATA_H @@ -27,7 +27,7 @@ #define LP_ORDER 16 ///< linear predictive coding filter order #define MIN_ISF_SPACING 50.0 // XXX: Taken from fixed-point 26.173, not sure #define PRED_FACTOR (1.0/3.0) -#define MIN_ENERGY -14.0 ///< initial innnovation energy (dB) +#define MIN_ENERGY -14.0 ///< initial innnovation energy (dB) #define ENERGY_MEAN 30.0 ///< mean innovation energy (dB) in all modes #define AMRWB_SUBFRAME_SIZE 64 ///< samples per subframe @@ -564,7 +564,7 @@ static const uint16_t order_MODE_23k05[] = { 7, AMR_OF(3, vq_gain), 6, 23, 45, 31, 35, 41, 27, 0 }; - + static const uint16_t order_MODE_23k85[] = { 1, AMR_BIT(vad), 0, 8, AMR_BIT(isp_id[0]), 48, 47, 9, 1, 8, 2, 7, 67, @@ -1810,5 +1810,5 @@ static const uint16_t cf_classA_size[] = { static const uint16_t cf_classB_size[] = { 78, 113, 181, 213, 245, 293, 325, 389, 405 }; - + #endif diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index e27ff6f..97c806e 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -47,22 +47,22 @@ typedef struct { float isf_q_past[LP_ORDER]; ///< quantized ISF vector of the previous frame double isp[4][LP_ORDER]; ///< ISP vectors from current frame double isp_sub4_past[LP_ORDER]; ///< ISP vector for the 4th subframe of the previous frame - + float lp_coef[4][LP_ORDER]; ///< Linear Prediction Coefficients from ISP vector uint8_t base_pitch_lag; ///< integer part of pitch lag for next relative subframe uint8_t pitch_lag_int; ///< integer part of pitch lag of the previous subframe - + float excitation_buf[PITCH_MAX + LP_ORDER + 1 + AMRWB_SUBFRAME_SIZE]; ///< current excitation and all necessary excitation history float *excitation; ///< points to current excitation in excitation_buf[] - + float pitch_vector[AMRWB_SUBFRAME_SIZE]; ///< adaptive codebook (pitch) vector for current subframe float fixed_vector[AMRWB_SUBFRAME_SIZE]; ///< algebraic codebook (fixed) vector - + float prediction_error[4]; ///< quantified prediction errors {20log10(^gamma_gc)} for previous four subframes float pitch_gain[5]; ///< quantified pitch gains for the current and previous four subframes float fixed_gain[5]; ///< quantified fixed gains for the current and previous four subframes - + float tilt_coef; ///< {beta_1} related to the voicing of the previous subframe float prev_sparse_fixed_gain; ///< previous fixed gain; used by anti-sparseness to determine "onset" @@ -75,29 +75,29 @@ typedef struct { float hpf_mem[4]; ///< previous values in the high-pass filter } AMRWBContext; -static int amrwb_decode_init(AVCodecContext *avctx) +static int amrwb_decode_init(AVCodecContext *avctx) { AMRWBContext *ctx = avctx->priv_data; int i; - + ctx->excitation = &ctx->excitation_buf[PITCH_MAX + LP_ORDER + 1]; for (i = 0; i < LP_ORDER; i++) { ctx->isf_q_past[i] = isf_init[i] / (float) (1 << 15); ctx->isp_sub4_past[i] = isp_init[i] / (float) (1 << 15); } - + ctx->tilt_coef = ctx->prev_tr_gain = 0.0; - + for (i = 0; i < 4; i++) ctx->prediction_error[i] = MIN_ENERGY; - + return 0; } /** * Parses a speech frame, storing data in the Context - * + * * @param c [in/out] the context * @param buf [in] pointer to the input buffer * @param buf_size [in] size of the input buffer @@ -110,30 +110,30 @@ static enum Mode unpack_bitstream(AMRWBContext *ctx, const uint8_t *buf, GetBitContext gb; enum Mode mode; uint16_t *data; - + init_get_bits(&gb, buf, buf_size * 8); - + /* AMR-WB header */ ctx->fr_cur_mode = get_bits(&gb, 4); mode = ctx->fr_cur_mode; ctx->fr_quality = get_bits1(&gb); - + skip_bits(&gb, 3); - + /* AMR-WB Auxiliary Information */ ctx->fr_mode_ind = get_bits(&gb, 4); ctx->fr_mode_req = get_bits(&gb, 4); // XXX: Need to check conformity in mode_ind/mode_req and crc? ctx->fr_crc = get_bits(&gb, 8); - + data = (uint16_t *) &ctx->frame; memset(data, 0, sizeof(AMRWBFrame)); buf++; - + if (mode < MODE_SID) { /* Normal speech frame */ const uint16_t *perm = amr_bit_orderings_by_mode[mode]; int field_size; - + while ((field_size = *perm++)) { int field = 0; int field_offset = *perm++; @@ -148,7 +148,7 @@ static enum Mode unpack_bitstream(AMRWBContext *ctx, const uint8_t *buf, else if (mode == MODE_SID) { /* Comfort noise frame */ /* not implemented */ } - + return mode; } @@ -167,8 +167,8 @@ static void isf2isp(const float *isf, double *isp) } /** - * Decodes quantized ISF vectors using 36-bit indices (6K60 mode only) - * + * Decodes quantized ISF vectors using 36-bit indices (6K60 mode only) + * * @param ind [in] array of 5 indices * @param isf_q [out] isf_q[LP_ORDER] * @param fr_q [in] frame quality (good frame == 1) @@ -176,7 +176,7 @@ static void isf2isp(const float *isf, double *isp) */ static void decode_isf_indices_36b(uint16_t *ind, float *isf_q, uint8_t fr_q) { int i; - + if (fr_q == 1) { for (i = 0; i < 9; i++) { isf_q[i] = dico1_isf[ind[0]][i] / (float) (1<<15); @@ -198,8 +198,8 @@ static void decode_isf_indices_36b(uint16_t *ind, float *isf_q, uint8_t fr_q) { } /** - * Decodes quantized ISF vectors using 46-bit indices (except 6K60 mode) - * + * Decodes quantized ISF vectors using 46-bit indices (except 6K60 mode) + * * @param ind [in] array of 7 indices * @param isf_q [out] isf_q[LP_ORDER] * @param fr_q [in] frame quality (good frame == 1) @@ -222,7 +222,7 @@ static void decode_isf_indices_46b(uint16_t *ind, float *isf_q, uint8_t fr_q) { isf_q[i + 3] = isf_q[i + 3] + dico22_isf[ind[3]][i] / (float) (1<<15); } for (i = 0; i < 3; i++) { - isf_q[i + 6] = isf_q[i + 6] + dico23_isf[ind[4]][i] / (float) (1<<15); + isf_q[i + 6] = isf_q[i + 6] + dico23_isf[ind[4]][i] / (float) (1<<15); } for (i = 0; i < 3; i++) { isf_q[i + 9] = isf_q[i + 9] + dico24_isf[ind[5]][i] / (float) (1<<15); @@ -237,7 +237,7 @@ static void decode_isf_indices_46b(uint16_t *ind, float *isf_q, uint8_t fr_q) { /** * Apply mean and past ISF values using the prediction factor * Updates past ISF vector - * + * * @param isf_q [in] current quantized ISF * @param isf_past [in/out] past quantized ISF * @@ -245,7 +245,7 @@ static void decode_isf_indices_46b(uint16_t *ind, float *isf_q, uint8_t fr_q) { static void isf_add_mean_and_past(float *isf_q, float *isf_past) { int i; float tmp; - + for (i = 0; i < LP_ORDER; i++) { tmp = isf_q[i]; isf_q[i] = tmp + isf_mean[i] / (float) (1<<15); @@ -256,7 +256,7 @@ static void isf_add_mean_and_past(float *isf_q, float *isf_past) { /** * Ensures a minimum distance between adjacent ISFs - * + * * @param isf [in/out] ISF vector * @param min_spacing [in] minimum gap to keep * @param size [in] ISF vector size @@ -265,7 +265,7 @@ static void isf_add_mean_and_past(float *isf_q, float *isf_past) { static void isf_set_min_dist(float *isf, float min_spacing, int size) { int i; float prev = 0.0; - + for (i = 0; i < size; i++) { isf[i] = FFMAX(isf[i], prev + min_spacing); prev = isf[i]; @@ -283,13 +283,13 @@ static void interpolate_isp(double isp_q[4][LP_ORDER], double *isp4_past) { int i; // XXX: Did not use ff_weighted_vector_sumf because using double - + for (i = 0; i < LP_ORDER; i++) isp_q[0][i] = 0.55 * isp4_past[i] + 0.45 * isp_q[3][i]; - + for (i = 0; i < LP_ORDER; i++) isp_q[1][i] = 0.20 * isp4_past[i] + 0.80 * isp_q[3][i]; - + for (i = 0; i < LP_ORDER; i++) isp_q[2][i] = 0.04 * isp4_past[i] + 0.96 * isp_q[3][i]; } @@ -308,20 +308,20 @@ static void isp2lp(double isp[LP_ORDER], float *lp, int lp_half_order) { double last_isp = isp[2 * lp_half_order - 1]; double qa_old = 0; // XXX: qa[i-2] assuming qa[-1] = 0, not mentioned in spec int i; - + ff_lsp2polyf(isp, pa, lp_half_order); ff_lsp2polyf(isp + 1, qa, lp_half_order); - + for (i=1; i<lp_half_order; i++) { double paf = (1 + last_isp) * pa[i]; double qaf = (1 - last_isp) * (qa[i] - qa_old); - + qa_old = qa[i-1]; - + lp[i] = 0.5 * (paf + qaf); lp2[i] = 0.5 * (paf - qaf); } - + lp2[0] = 0.5 * (1 + last_isp) * pa[lp_half_order] * lp_half_order; lp2[lp_half_order] = last_isp; } @@ -416,7 +416,7 @@ static void decode_pitch_vector(AMRWBContext *ctx, int i; float *exc = ctx->excitation; enum Mode mode = ctx->fr_cur_mode; - + if (mode == MODE_6k60) { decode_pitch_lag_6K60(&pitch_lag_int, &pitch_lag_frac, amr_subframe->adap, &ctx->base_pitch_lag, subframe); @@ -426,10 +426,10 @@ static void decode_pitch_vector(AMRWBContext *ctx, } else decode_pitch_lag_high(&pitch_lag_int, &pitch_lag_frac, amr_subframe->adap, &ctx->base_pitch_lag, subframe); - + ctx->pitch_lag_int = pitch_lag_int; pitch_lag_int += pitch_lag_frac > 0; - + /* Calculate the pitch vector by interpolating the past excitation at the pitch lag using a hamming windowed sinc function. */ ff_acelp_interpolatef(exc, exc + 1 - pitch_lag_int, @@ -464,7 +464,7 @@ static void decode_pitch_vector(AMRWBContext *ctx, static inline void decode_1p_track(int *out, int code, int m, int off) { ///code: m+1 bits int pos = BIT_STR(code, 0, m) + off; - + out[0] = BIT_POS(code, m) ? -pos : pos; } @@ -472,7 +472,7 @@ static inline void decode_2p_track(int *out, int code, int m, int off) { ///code: 2m+1 bits int pos0 = BIT_STR(code, m, m) + off; int pos1 = BIT_STR(code, 0, m) + off; - + out[0] = BIT_POS(code, 2*m) ? -pos0 : pos0; out[1] = BIT_POS(code, 2*m) ? -pos1 : pos1; out[1] = pos0 > pos1 ? -out[1] : out[1]; @@ -481,7 +481,7 @@ static inline void decode_2p_track(int *out, int code, int m, int off) static void decode_3p_track(int *out, int code, int m, int off) { ///code: 3m+1 bits int half_2p = BIT_POS(code, 2*m - 1) << (m - 1); - + decode_2p_track(out, BIT_STR(code, 0, 2*m - 1), m - 1, off + half_2p); decode_1p_track(out + 2, BIT_STR(code, 2*m, m + 1), m, off); @@ -491,13 +491,13 @@ static void decode_4p_track(int *out, int code, int m, int off) { ///code: 4m bits int half_4p, subhalf_2p; int b_offset = 1 << (m - 1); - + switch (BIT_STR(code, 4*m - 2, 2)) /* case ID (2 bits) */ { case 0: /* 0 pulses in A, 4 pulses in B or vice-versa */ half_4p = BIT_POS(code, 4*m - 3) << (m - 1); /* which has 4 pulses */ subhalf_2p = BIT_POS(code, 2*m - 3) << (m - 2); - + decode_2p_track(out, BIT_STR(code, 0, 2*m - 3), m - 2, off + half_4p + subhalf_2p); decode_2p_track(out + 2, BIT_STR(code, 2*m - 2, 2*m - 1), @@ -527,7 +527,7 @@ static void decode_4p_track(int *out, int code, int m, int off) static void decode_5p_track(int *out, int code, int m, int off) { ///code: 5m bits int half_3p = BIT_POS(code, 5*m - 1) << (m - 1); - + decode_3p_track(out, BIT_STR(code, 2*m, 3*m - 2), m - 1, off + half_3p); // XXX: there seems to be a typo in I3p expoent (from reference) @@ -540,7 +540,7 @@ static void decode_6p_track(int *out, int code, int m, int off) /* which half has more pulses in cases 0 to 2 */ int half_more = BIT_POS(code, 6*m - 5) << (m - 1); int half_other = b_offset - half_more; - + switch (BIT_STR(code, 6*m - 4, 2)) /* case ID (2 bits) */ { case 0: /* 0 pulses in A, 6 pulses in B or vice-versa */ @@ -590,7 +590,7 @@ static void decode_fixed_sparse(AMRFixed *fixed_sparse, const uint16_t *pulse_hi int pulses_nb = 0; int spacing = (mode == MODE_6k60) ? 2 : 4; int i, j; - + switch (mode) { case MODE_6k60: for (i = 0; i < 2; i++) @@ -642,7 +642,7 @@ static void decode_fixed_sparse(AMRFixed *fixed_sparse, const uint16_t *pulse_hi fixed_sparse->y[pulses_nb] = pos < 0 ? -1.0 : 1.0; pulses_nb++; } - + fixed_sparse->n = pulses_nb; } @@ -658,12 +658,12 @@ static void decode_gains(const uint8_t vq_gain, const enum Mode mode, float *fixed_gain_factor, float *pitch_gain) { const int16_t *gains; - + if (mode == MODE_6k60 || mode == MODE_8k85) gains = qua_gain_6b[vq_gain]; else gains = qua_gain_7b[vq_gain]; - + *pitch_gain = gains[0] * (1.0 / 16384.0); *fixed_gain_factor = gains[1] * (1.0 / 2048.0); } @@ -684,10 +684,10 @@ static void pitch_sharpening(AMRWBContext *ctx, AMRFixed *fixed_sparse, /* Periodicity enhancement part */ fixed_sparse->pitch_lag = ctx->pitch_lag_int; fixed_sparse->pitch_fac = 0.85; - + ff_set_fixed_vector(fixed_vector, fixed_sparse, 1.0, AMRWB_SUBFRAME_SIZE); - + /* Tilt part */ ff_weighted_vector_sumf(fixed_vector + 1, fixed_vector + 1, fixed_vector, 1.0, - ctx->tilt_coef, AMRWB_SUBFRAME_SIZE - 1); @@ -990,9 +990,9 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, float synth_fixed_gain; // the fixed gain that synthesis should use float voice_fac, stab_fac; // parameters used for gain smoothing int sub, i; - + ctx->fr_cur_mode = unpack_bitstream(ctx, buf, buf_size); - + if (ctx->fr_cur_mode == MODE_SID) { av_log_missing_feature(avctx, "SID mode", 1); return -1; @@ -1000,7 +1000,7 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, if (!ctx->fr_quality) { av_log(avctx, AV_LOG_ERROR, "Encountered a bad or corrupted frame\n"); } - + /* Decode the quantized ISF vector */ if (ctx->fr_cur_mode == MODE_6k60) { decode_isf_indices_36b(cf->isp_id, ctx->isf_quant, ctx->fr_quality); @@ -1010,17 +1010,17 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, } stab_fac = stability_factor(ctx->isf_quant, ctx->isf_q_past); - + isf_add_mean_and_past(ctx->isf_quant, ctx->isf_q_past); isf_set_min_dist(ctx->isf_quant, MIN_ISF_SPACING, LP_ORDER); - + isf2isp(ctx->isf_quant, ctx->isp[3]); /* Generate a ISP vector for each subframe */ interpolate_isp(ctx->isp, ctx->isp_sub4_past); - + for (sub = 0; sub < 4; sub++) isp2lp(ctx->isp[sub], ctx->lp_coef[sub], LP_ORDER/2); - + for (sub = 0; sub < 4; sub++) { const AMRWBSubFrame *cur_subframe = &cf->subframe[sub]; @@ -1032,9 +1032,9 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, decode_gains(cur_subframe->vq_gain, ctx->fr_cur_mode, &fixed_gain_factor, &ctx->pitch_gain[4]); - + pitch_sharpening(ctx, &fixed_sparse, ctx->fixed_vector); - + ctx->fixed_gain[4] = ff_amr_set_fixed_gain(fixed_gain_factor, ff_dot_productf(ctx->fixed_vector, ctx->fixed_vector, @@ -1046,7 +1046,7 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, voice_fac = voice_factor(ctx->pitch_vector, ctx->pitch_gain[4], ctx->fixed_vector, ctx->fixed_gain[4]); ctx->tilt_coef = voice_fac * 0.25 + 0.25; - + /* Construct current excitation */ for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) { ctx->excitation[i] *= ctx->pitch_gain[4]; @@ -1060,7 +1060,7 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, /* Post-processing of excitation elements */ synth_fixed_gain = noise_enhancer(ctx->fixed_gain[4], &ctx->prev_tr_gain, voice_fac, stab_fac); - + synth_fixed_vector = anti_sparseness(ctx, ctx->fixed_vector, spare_vector); @@ -1085,10 +1085,10 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, AMRWB_SUBFRAME_SIZE); update_sub_state(ctx); } - + // update state for next frame memcpy(ctx->isp_sub4_past, ctx->isp[3], LP_ORDER * sizeof(ctx->isp[3][0])); - + return 0; } commit 89dcf67e516dc9b0b04f890ab8a22b136f6163f7 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Sun Jul 18 00:50:11 2010 -0300 Write the high-pass filter for synthesized speech diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 442944d..8f1d027 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -1789,6 +1789,13 @@ static const float *ir_filters_lookup[2] = { ir_filter_str, ir_filter_mid }; +/* High-pass filter coefficients for inputs and outputs (feedback) */ +static const float hpf_x_coef[3] = { + 0.989501953, -1.979003906, 0.989501953 +}; + +static const float hpf_y_coef[2] = { 1.978881836, -0.979125977 }; + /* Core frame sizes in each mode */ static const uint16_t cf_sizes_wb[] = { 132, 177, 253, 285, 317, 365, 397, 461, 477, diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 95aebfb..e27ff6f 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -72,6 +72,7 @@ typedef struct { float samples_in[LP_ORDER + AMRWB_SUBFRAME_SIZE]; ///< floating point samples float demph_mem[1]; ///< previous value in the de-emphasis filter + float hpf_mem[4]; ///< previous values in the high-pass filter } AMRWBContext; static int amrwb_decode_init(AVCodecContext *avctx) @@ -937,6 +938,36 @@ static void de_emphasis(float *synth, float m, float mem[1]) } /** + * Apply to synthesis a 2nd order high-pass filter + * with cutoff frequency at 31 Hz + * + * @param out [out] buffer for filtered output + * @param mem [in] state from last filtering + * @param in [in] input speech data + * + * @remark It is safe to pass the same array in in and out parameters. + */ +static void high_pass_filter(float *out, float mem[4], const float *in) +{ + int i; + float *x = mem - 1, *y = mem + 2; // previous inputs and outputs + + for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) { + float x0 = in[i]; + + out[i] = hpf_x_coef[0] * x0 + hpf_y_coef[0] * y[0] + + hpf_x_coef[1] * x[1] + hpf_y_coef[1] * y[1] + + hpf_x_coef[2] * x[2]; + + y[1] = y[0]; + y[0] = out[i]; + + x[2] = x[1]; + x[1] = x0; + } +} + +/** * Update context state before the next subframe */ static void update_sub_state(AMRWBContext *ctx) @@ -1043,8 +1074,12 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, synthesis(ctx, ctx->lp_coef[sub], synth_fixed_gain, synth_fixed_vector, &ctx->samples_in[LP_ORDER], 1); + /* Synthesis speech post-processing */ de_emphasis(&ctx->samples_in[LP_ORDER], PREEMPH_FAC, ctx->demph_mem); + high_pass_filter(&ctx->samples_in[LP_ORDER], ctx->hpf_mem, + &ctx->samples_in[LP_ORDER]); + /* Update buffers and history */ ff_clear_fixed_vector(ctx->fixed_vector, &fixed_sparse, AMRWB_SUBFRAME_SIZE); commit 3c771311f0281d2858c9f53636011069b6e202db Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Sat Jul 17 23:36:38 2010 -0300 Write de-emphasis filter for synthesized speech diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 06e412c..442944d 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -34,7 +34,7 @@ #define AMRWB_SAMPLE_BOUND 32768.0 ///< threshold for synthesis overflow #define PITCH_MAX 231 ///< maximum received pitch delay value - +#define PREEMPH_FAC 0.68 ///< factor used to de-emphasize synthesis /* Mode ordering is sensitive, do not change */ enum Mode { diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index fa86de4..95aebfb 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -70,6 +70,8 @@ typedef struct { float prev_tr_gain; ///< previous initial gain used by noise enhancer for thresold float samples_in[LP_ORDER + AMRWB_SUBFRAME_SIZE]; ///< floating point samples + + float demph_mem[1]; ///< previous value in the de-emphasis filter } AMRWBContext; static int amrwb_decode_init(AVCodecContext *avctx) @@ -915,6 +917,26 @@ static uint8_t synthesis(AMRWBContext *ctx, float *lpc, } /** + * Apply to synthesis a de-emphasis filter of the form: + * H(z) = 1 / (1 - m * z^-1) + * + * @param synth [in/out] synthesized speech array + * @param m [in] filter coefficient + * @param mem [in] state from last filtering + */ +static void de_emphasis(float *synth, float m, float mem[1]) +{ + int i; + + synth[0] += m * mem[0]; + + for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) + synth[i] += synth[i - 1] * m; + + mem[0] = synth[AMRWB_SUBFRAME_SIZE - 1]; +} + +/** * Update context state before the next subframe */ static void update_sub_state(AMRWBContext *ctx) @@ -1021,6 +1043,8 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, synthesis(ctx, ctx->lp_coef[sub], synth_fixed_gain, synth_fixed_vector, &ctx->samples_in[LP_ORDER], 1); + de_emphasis(&ctx->samples_in[LP_ORDER], PREEMPH_FAC, ctx->demph_mem); + /* Update buffers and history */ ff_clear_fixed_vector(ctx->fixed_vector, &fixed_sparse, AMRWB_SUBFRAME_SIZE); ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdata.h | 17 +++- libavcodec/amrwbdec.c | 191 +++++++++++++++++++++++++++++++----------------- 2 files changed, 137 insertions(+), 71 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0

[soc]: r5863 - in libavfilter: Makefile allfilters.c diffs/01_build.diff
by stefano 17 Jul '10

17 Jul '10

Author: stefano Date: Sat Jul 17 22:24:32 2010 New Revision: 5863 Log: Do not literally copy the allfilters.c and Makefile files anymore, apply the patch 01_build.diff instead. Simplify the integration work a bit. Added: libavfilter/diffs/01_build.diff Deleted: libavfilter/Makefile libavfilter/allfilters.c Added: libavfilter/diffs/01_build.diff ============================================================================== --- /dev/null 00:00:00 1970 (empty, because file is newly added) +++ libavfilter/diffs/01_build.diff Sat Jul 17 22:24:32 2010 (r5863) @@ -0,0 +1,88 @@ +Index: libavfilter/Makefile +=================================================================== +--- libavfilter/Makefile (revision 24294) ++++ libavfilter/Makefile (working copy) +@@ -1,10 +1,11 @@ + include $(SUBDIR)../config.mak + + NAME = avfilter +-FFLIBS = avcodec avutil swscale ++FFLIBS = avcodec avutil ++FFLIBS-$(CONFIG_SCALE_FILTER) += swscale + FFLIBS-$(CONFIG_AVFILTER_LAVF) += avformat + +-HEADERS = avfilter.h ++HEADERS = avfilter.h avfiltergraph.h + + OBJS = allfilters.o \ + avfilter.o \ +@@ -16,19 +17,32 @@ + + OBJS-$(CONFIG_ASPECT_FILTER) += vf_aspect.o + OBJS-$(CONFIG_CROP_FILTER) += vf_crop.o ++OBJS-$(CONFIG_DRAWBOX_FILTER) += vf_drawbox.o ++OBJS-$(CONFIG_DRAWTEXT_FILTER) += vf_drawtext.o ++OBJS-$(CONFIG_FADE_FILTER) += vf_fade.o ++OBJS-$(CONFIG_FIFO_FILTER) += vf_fifo.o + OBJS-$(CONFIG_FORMAT_FILTER) += vf_format.o ++OBJS-$(CONFIG_FPS_FILTER) += vf_fps.o ++OBJS-$(CONFIG_HFLIP_FILTER) += vf_hflip.o ++OBJS-$(CONFIG_NEGATE_FILTER) += vf_negate.o + OBJS-$(CONFIG_NOFORMAT_FILTER) += vf_format.o + OBJS-$(CONFIG_NULL_FILTER) += vf_null.o ++OBJS-$(CONFIG_OVERLAY_FILTER) += vf_overlay.o + OBJS-$(CONFIG_PAD_FILTER) += vf_pad.o + OBJS-$(CONFIG_PIXDESCTEST_FILTER) += vf_pixdesctest.o + OBJS-$(CONFIG_PIXELASPECT_FILTER) += vf_aspect.o ++OBJS-$(CONFIG_ROTATE_FILTER) += vf_rotate.o + OBJS-$(CONFIG_SCALE_FILTER) += vf_scale.o ++OBJS-$(CONFIG_SETPTS_FILTER) += vf_setpts.o + OBJS-$(CONFIG_SLICIFY_FILTER) += vf_slicify.o ++OBJS-$(CONFIG_SPLIT_FILTER) += vf_split.o ++OBJS-$(CONFIG_TRANSPOSE_FILTER) += vf_transpose.o + OBJS-$(CONFIG_UNSHARP_FILTER) += vf_unsharp.o + OBJS-$(CONFIG_VFLIP_FILTER) += vf_vflip.o + + OBJS-$(CONFIG_BUFFER_FILTER) += vsrc_buffer.o + OBJS-$(CONFIG_COLOR_FILTER) += vf_pad.o ++OBJS-$(CONFIG_MOVIE_FILTER) += vsrc_movie.o + OBJS-$(CONFIG_NULLSRC_FILTER) += vsrc_nullsrc.o + + OBJS-$(CONFIG_NULLSINK_FILTER) += vsink_nullsink.o +Index: libavfilter/allfilters.c +=================================================================== +--- libavfilter/allfilters.c (revision 24294) ++++ libavfilter/allfilters.c (working copy) +@@ -36,19 +36,32 @@ + + REGISTER_FILTER (ASPECT, aspect, vf); + REGISTER_FILTER (CROP, crop, vf); ++ REGISTER_FILTER (DRAWBOX, drawbox, vf); ++ REGISTER_FILTER (DRAWTEXT, drawtext, vf); ++ REGISTER_FILTER (FADE, fade, vf); ++ REGISTER_FILTER (FIFO, fifo, vf); + REGISTER_FILTER (FORMAT, format, vf); ++ REGISTER_FILTER (FPS, fps, vf); ++ REGISTER_FILTER (HFLIP, hflip, vf); ++ REGISTER_FILTER (NEGATE, negate, vf); + REGISTER_FILTER (NOFORMAT, noformat, vf); + REGISTER_FILTER (NULL, null, vf); ++ REGISTER_FILTER (OVERLAY, overlay, vf); + REGISTER_FILTER (PAD, pad, vf); + REGISTER_FILTER (PIXDESCTEST, pixdesctest, vf); + REGISTER_FILTER (PIXELASPECT, pixelaspect, vf); ++ REGISTER_FILTER (ROTATE, rotate, vf); + REGISTER_FILTER (SCALE, scale, vf); ++ REGISTER_FILTER (SETPTS, setpts, vf); + REGISTER_FILTER (SLICIFY, slicify, vf); ++ REGISTER_FILTER (SPLIT, split, vf); ++ REGISTER_FILTER (TRANSPOSE, transpose, vf); + REGISTER_FILTER (UNSHARP, unsharp, vf); + REGISTER_FILTER (VFLIP, vflip, vf); + + REGISTER_FILTER (BUFFER, buffer, vsrc); + REGISTER_FILTER (COLOR, color, vsrc); ++ REGISTER_FILTER (MOVIE, movie, vsrc); + REGISTER_FILTER (NULLSRC, nullsrc, vsrc); + + REGISTER_FILTER (NULLSINK, nullsink, vsink);

1 0

[soc]: r5862 - in libavfilter: Makefile allfilters.c checkout.sh diffs/00_configure.diff diffs/03_doc_filters.diff
by stefano 17 Jul '10

17 Jul '10

Author: stefano Date: Sat Jul 17 22:16:21 2010 New Revision: 5862 Log: Update to FFmpeg r24294 / libswscale r31746. Modified: libavfilter/Makefile libavfilter/allfilters.c libavfilter/checkout.sh libavfilter/diffs/00_configure.diff libavfilter/diffs/03_doc_filters.diff Modified: libavfilter/Makefile ============================================================================== --- libavfilter/Makefile Thu Jul 15 18:34:29 2010 (r5861) +++ libavfilter/Makefile Sat Jul 17 22:16:21 2010 (r5862) @@ -29,6 +29,7 @@ OBJS-$(CONFIG_NOFORMAT_FILTER) OBJS-$(CONFIG_NULL_FILTER) += vf_null.o OBJS-$(CONFIG_OVERLAY_FILTER) += vf_overlay.o OBJS-$(CONFIG_PAD_FILTER) += vf_pad.o +OBJS-$(CONFIG_PIXDESCTEST_FILTER) += vf_pixdesctest.o OBJS-$(CONFIG_PIXELASPECT_FILTER) += vf_aspect.o OBJS-$(CONFIG_ROTATE_FILTER) += vf_rotate.o OBJS-$(CONFIG_SCALE_FILTER) += vf_scale.o @@ -40,6 +41,7 @@ OBJS-$(CONFIG_UNSHARP_FILTER) OBJS-$(CONFIG_VFLIP_FILTER) += vf_vflip.o OBJS-$(CONFIG_BUFFER_FILTER) += vsrc_buffer.o +OBJS-$(CONFIG_COLOR_FILTER) += vf_pad.o OBJS-$(CONFIG_MOVIE_FILTER) += vsrc_movie.o OBJS-$(CONFIG_NULLSRC_FILTER) += vsrc_nullsrc.o Modified: libavfilter/allfilters.c ============================================================================== --- libavfilter/allfilters.c Thu Jul 15 18:34:29 2010 (r5861) +++ libavfilter/allfilters.c Sat Jul 17 22:16:21 2010 (r5862) @@ -48,6 +48,7 @@ void avfilter_register_all(void) REGISTER_FILTER (NULL, null, vf); REGISTER_FILTER (OVERLAY, overlay, vf); REGISTER_FILTER (PAD, pad, vf); + REGISTER_FILTER (PIXDESCTEST, pixdesctest, vf); REGISTER_FILTER (PIXELASPECT, pixelaspect, vf); REGISTER_FILTER (ROTATE, rotate, vf); REGISTER_FILTER (SCALE, scale, vf); @@ -59,6 +60,7 @@ void avfilter_register_all(void) REGISTER_FILTER (VFLIP, vflip, vf); REGISTER_FILTER (BUFFER, buffer, vsrc); + REGISTER_FILTER (COLOR, color, vsrc); REGISTER_FILTER (MOVIE, movie, vsrc); REGISTER_FILTER (NULLSRC, nullsrc, vsrc); Modified: libavfilter/checkout.sh ============================================================================== --- libavfilter/checkout.sh Thu Jul 15 18:34:29 2010 (r5861) +++ libavfilter/checkout.sh Sat Jul 17 22:16:21 2010 (r5862) @@ -1,11 +1,11 @@ #! /bin/sh echo "checking out pristine ffmpeg" -svn checkout svn://svn.ffmpeg.org/ffmpeg/trunk/ ffmpeg -r24040 +svn checkout svn://svn.ffmpeg.org/ffmpeg/trunk/ ffmpeg -r24294 echo "downloading the corresponding version of swscale" cd ffmpeg/libswscale -svn up -r31628 +svn up -r31746 cd ../.. echo "patching ffmpeg" Modified: libavfilter/diffs/00_configure.diff ============================================================================== --- libavfilter/diffs/00_configure.diff Thu Jul 15 18:34:29 2010 (r5861) +++ libavfilter/diffs/00_configure.diff Sat Jul 17 22:16:21 2010 (r5862) @@ -1,6 +1,6 @@ Index: configure =================================================================== ---- configure (revision 24040) +--- configure (revision 24294) +++ configure (working copy) @@ -167,6 +167,7 @@ and libraw1394 [no] @@ -10,7 +10,7 @@ Index: configure --enable-libgsm enable GSM support via libgsm [no] --enable-libmp3lame enable MP3 encoding via libmp3lame [no] --enable-libnut enable NUT (de)muxing via libnut, -@@ -860,6 +861,7 @@ +@@ -869,6 +870,7 @@ libdc1394 libdirac libfaac @@ -18,7 +18,7 @@ Index: configure libgsm libmp3lame libnut -@@ -1001,6 +1003,7 @@ +@@ -1010,6 +1012,7 @@ llrintf local_aligned_16 local_aligned_8 @@ -26,7 +26,7 @@ Index: configure log2 log2f loongson -@@ -1369,6 +1372,7 @@ +@@ -1376,6 +1379,7 @@ udp_protocol_deps="network" # filters @@ -34,7 +34,7 @@ Index: configure movie_filter_deps="avfilter_lavf" avfilter_lavf_deps="avformat" -@@ -2461,6 +2465,7 @@ +@@ -2559,6 +2563,7 @@ check_struct "sys/time.h sys/resource.h" "struct rusage" ru_maxrss check_func inet_aton $network_extralibs check_func isatty @@ -42,7 +42,7 @@ Index: configure check_func ${malloc_prefix}memalign && enable memalign check_func mkstemp check_func ${malloc_prefix}posix_memalign && enable posix_memalign -@@ -2543,6 +2548,7 @@ +@@ -2641,6 +2646,7 @@ require libdirac libdirac_decoder/dirac_parser.h dirac_decoder_init $(pkg-config --libs dirac) && require libdirac libdirac_encoder/dirac_encoder.h dirac_encoder_init $(pkg-config --libs dirac) enabled libfaac && require2 libfaac "stdint.h faac.h" faacEncGetVersion -lfaac Modified: libavfilter/diffs/03_doc_filters.diff ============================================================================== --- libavfilter/diffs/03_doc_filters.diff Thu Jul 15 18:34:29 2010 (r5861) +++ libavfilter/diffs/03_doc_filters.diff Sat Jul 17 22:16:21 2010 (r5862) @@ -1,6 +1,6 @@ Index: doc/filters.texi =================================================================== ---- doc/filters.texi (revision 24040) +--- doc/filters.texi (revision 24294) +++ doc/filters.texi (working copy) @@ -39,6 +39,114 @@ @@ -205,9 +205,9 @@ Index: doc/filters.texi @section pad Add paddings to the input image, and places the original input at the -@@ -112,6 +294,14 @@ +@@ -124,6 +306,14 @@ - @end table + can be used to test the monowhite pixel format descriptor definition. +@section rotate + @@ -220,7 +220,7 @@ Index: doc/filters.texi @section scale Scale the input video to @var{width}:@var{height} and/or convert the image format. -@@ -137,6 +327,24 @@ +@@ -149,6 +339,24 @@ The default value of @var{width} and @var{height} is 0. @@ -245,7 +245,7 @@ Index: doc/filters.texi @section slicify Pass the images of input video on to next video filter as multiple -@@ -152,6 +360,19 @@ +@@ -164,6 +372,19 @@ Adding this in the beginning of filter chains should make filtering faster due to better use of the memory cache. @@ -265,7 +265,7 @@ Index: doc/filters.texi @section unsharp Sharpen or blur the input video. -@@ -217,6 +438,32 @@ +@@ -229,6 +450,11 @@ Below is a description of the currently available video sources. @@ -274,6 +274,13 @@ Index: doc/filters.texi +This input filter is used by the client application to feed pictures to +the filter chain. See ffmpeg.c for an usage example. + + @section color + + Provide an uniformly colored input. +@@ -268,6 +494,27 @@ + "color=red@@0.2:qcif:10 [color]; [in][color] overlay [out]" + @end example + +@section movie + +The parameters of the movie filter are

1 0