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

[PATCH] xiph packetizer
by Josh Allmann 07 Aug '10

07 Aug '10

Hi guys, Here is a first attempt at a RTP packetizer for the Xiph codecs. This is still a work in progress; Vorbis likes to crash when rescaling the timebase, and the Theora output only vaguely resembles what it should. (There are some unrelated changes in there, particularly to the depacketizer. I will separate those later.) The only major architectural issue here is that the configuration identifier (here hard-coded to 0xfedcba) needs to be carried across from the SDP to the packetizer proper. Right now I don't believe there is a clean way to do that without adding another field to RTPMuxContext, a generic priv_data or something. The "best" way, IMO, would be to structure RTP packetization in a way similar to how depacketization is done -- through dynamic handlers. We might be able to get away with a hard-coded configuration identifier, but I don't know yet if my samples will periodically emit in-band extradata. (The RFC provisions for this.) In that case, the config ID needs to be refreshed, and kept across subsequent packets. Josh

4 31

[soc] Questions on patch order, threads and review comments
by S.N. Hemanth Meenakshisundaram 04 Aug '10

04 Aug '10

Hi Stefano, I have a few questions on the order of patches and on some of your review comments from earlier: >> > >> >Here is the working af_resample.c and associated Makefile and >> >allfilters.c changes. >> > >> >[...] >> >> > >> -#define AV_PERM_READ 0x01 ///< can read from the buffer >> -#define AV_PERM_WRITE 0x02 ///< can write to the buffer >> -#define AV_PERM_PRESERVE 0x04 ///< nobody else can overwrite the >> buffer >> -#define AV_PERM_REUSE 0x08 ///< can output the buffer multiple >> times, with the same contents each time >> -#define AV_PERM_REUSE2 0x10 ///< can output the buffer multiple >> times, modified each time > > The AV_PERM_* moving can go to a separate patch. Should I submit the AV_PERM_* patch first (move it out of AVFilterPicRef structure) and then the the audio framework changes patch? Or should I focus on the audio filter patches and define the PERM macros twice (inside AVFilterPicRef and AVFilterSamplesRef structures). > >> - link->format = link->in_formats->formats[0]; >> + if (link->type == AVMEDIA_TYPE_VIDEO) >> + link->format = link->in_formats->formats[0]; >> + else if (link->type == AVMEDIA_TYPE_AUDIO) >> + link->aformat = link->in_formats->aformats[0]; > > simpler: link->format = type == VIDEO ? link->in_formats->formats[0] : > link->in_formats->aformats[0]; Here I need to assign to link->format in case of video and link->aformat in case of audio. How can I get this done using the ?: form above? > >> +#define CMP_AND_ADD(acount, bcount, afmt, bfmt, retfmt) { \ >> + for(i = 0; i < acount; i++) \ >> + for(j = 0; j < bcount; j++) \ >> + if(afmt[i] == bfmt[j]) \ >> + retfmt[k++] = afmt[i]; \ >> +} >> [...] >> > > This refactoring is nice but please split it in a separate patch (and > separate thread). Again, should I submit the CMP_AND_ADD refactoring patch first before the audio filter patches? I should start a new thread for each patch - is that right? Also, I now have some patches based on your comments and some of the additions for channel layout conversion etc. Should I send these as patches of patches or send the entire patch each time starting from whatever is not yet in the soc svns? Patches of patches would be difficult to read while sending the entire patch again hides the new changes. Finally, Wikipedia says the downmix formulae for ac3 to stereo are: Left Only = Left (Front) + -3dB*Center + att*LeftSide Right Only = Right (Front) + -3dB*Center + att*RightSide where att = -3dB, -6dB or -9dB. The article has no references. Can I just go ahead and use this? It seems to work well for a 5.0 track I have with att = -9dB. Regards, Hemanth Regards, Hemanth

4 5

Re: [FFmpeg-soc] [PATCH] Make avfi graph call resample filter for filters with conflicting sample fmts.
by S.N. Hemanth Meenakshisundaram 04 Aug '10

04 Aug '10

If two filters have incompatible sample formats, attempt inserting a scaler (resample filter). --- libavfilter/avfiltergraph.c | 14 ++++++++++---- 1 files changed, 10 insertions(+), 4 deletions(-)

2 1

Fwd: [PATCH] ffmpeg.c changes for calling lavfi audio framework
by S.N. Hemanth Meenakshisundaram 31 Jul '10

31 Jul '10

ffmpeg changes to call lavfi audio framework. It works but is a temporary way to test the other audio filters. Couple if issues with this: 1. Right now, asrc and ffmpeg.c copy any incoming audio frames and thewn copy them back out. To avoid the two memcpy calls. is it ok to implement interfaces to avoid this? Please let me know. I was thinking of defining alternative get_video_buffer and unref_samples to pass in a buffer pointer alone and use this as input parameter. Comments? Regards, --- ffmpeg.c | 182 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-- 1 files changed, 177 insertions(+), 5 deletions(-)

2 2

Re: [FFmpeg-soc] [PATCH] fixes for asrc_buffer
by S.N. Hemanth Meenakshisundaram 31 Jul '10

31 Jul '10

Some fixes to the asrc_buffer. Add config props since src filter has no inputs and default config props will fail. --- libavfilter/allfilters.c | 2 +- libavfilter/asrc_buffer.c | 10 ++++++++-- 2 files changed, 9 insertions(+), 3 deletions(-)

2 3

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 30 Jul '10

30 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 121492a18969d4665210fe6dede43482776acce8 (commit) via 8cbd44b44a1b8d4332fc8d22ced69a93d85a31fc (commit) from 01100290fe3b220e4fbbe6ec16f7f85652d25092 (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 121492a18969d4665210fe6dede43482776acce8 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Fri Jul 30 02:34:33 2010 -0300 Write post-processing FIR filters for the high-band diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 76a1a60..17f7d9e 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -28,6 +28,7 @@ #define LP_ORDER_16k 20 ///< lpc filter order at 16kHz #define UPS_FIR_SIZE 12 ///< upsampling filter size #define UPS_MEM_SIZE 2 * UPS_FIR_SIZE +#define HB_FIR_SIZE 30 ///< amount of past data needed by HB filters #define MIN_ISF_SPACING (128 / 32768.0) ///< minimum isf gap #define PRED_FACTOR (1.0 / 3.0) @@ -1840,6 +1841,36 @@ static const uint16_t qua_hb_gain[16] = { 11210, 12206, 13391, 14844, 16770, 19655, 24289, 32728 }; +/* High-band post-processing FIR filters coefficients in Q15 */ +// XXX: not sure if it is Q15 indeed (guessing) +static const float bpf_6_7_coef[31] = { // band pass, 6kHz and 7kHz cutoffs + -9.765625e-04, 1.434326e-03, 9.765625e-04, + -8.239746e-04, -1.126099e-02, 3.424072e-02, + -4.336548e-02, 0.000000e+00, 1.159058e-01, + -2.709961e-01, 3.768616e-01, -3.352051e-01, + 1.082764e-01, 2.369995e-01, -5.493469e-01, + 6.749878e-01, + -5.493469e-01, 2.369995e-01, 1.082764e-01, + -3.352051e-01, 3.768616e-01, -2.709961e-01, + 1.159058e-01, 0.000000e+00, -4.336548e-02, + 3.424072e-02, -1.126099e-02, -8.239746e-04, + 9.765625e-04, 1.434326e-03, -9.765625e-04 +}; + +static const float lpf_7_coef[31] = { // low pass, 7kHz cutoff + -6.408691e-04, 1.434326e-03, -2.716064e-03, + 4.455566e-03, -6.195068e-03, 6.988525e-03, + -5.401611e-03, 0.000000e+00, 1.022339e-02, + -2.560425e-02, 4.531860e-02, -6.747437e-02, + 8.944702e-02, -1.080933e-01, 1.206360e-01, + 8.753052e-01, + 1.206360e-01, -1.080933e-01, 8.944702e-02, + -6.747437e-02, 4.531860e-02, -2.560425e-02, + 1.022339e-02, 0.000000e+00, -5.401611e-03, + 6.988525e-03, -6.195068e-03, 4.455566e-03, + -2.716064e-03, 1.434326e-03, -6.408691e-04 +}; + /* 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 2786816..de4b399 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -73,11 +73,13 @@ typedef struct { float prev_tr_gain; ///< previous initial gain used by noise enhancer for thresold float samples_az[LP_ORDER + AMRWB_SUBFRAME_SIZE]; ///< lower band samples from synthesis at 12.8kHz - float samples_up[UPS_MEM_SIZE + AMRWB_SUBFRAME_SIZE]; ///< lower band samples processed for upsampling at 12.8kHz + float samples_up[UPS_MEM_SIZE + AMRWB_SUBFRAME_SIZE]; ///< lower band samples processed for upsampling float samples_hb[LP_ORDER_16k + AMRWB_SFR_SIZE_OUT]; ///< higher band samples from synthesis at 16kHz - float hpf_31_mem[4], hpf_400_mem[4]; ///< previous values in the high-pass filters + float hpf_31_mem[4], hpf_400_mem[4]; ///< previous values in the high pass filters float demph_mem[1]; ///< previous value in the de-emphasis filter + float bpf_6_7_mem[HB_FIR_SIZE]; ///< previous values in the high-band band pass filter + float lpf_7_mem[HB_FIR_SIZE]; ///< previous values in the high-band low pass filter AVLFG prng; ///< random number generator for white noise excitation uint8_t first_frame; ///< flag active during decoding of the first frame @@ -1176,6 +1178,39 @@ static void hb_synthesis(AMRWBContext *ctx, int subframe, float *samples, } /** + * Apply to high-band samples a 15th order filter + * The filter characteristic depends on the given coefficients + * + * @param out [out] buffer for filtered output + * @param fir_coef [in] filter coefficients + * @param mem [in/out] state from last filtering (updated) + * @param cp_gain [in] compensation gain (usually the filter gain) + * @param in [in] input speech data (high-band) + * + * @remark It is safe to pass the same array in in and out parameters. + */ +static void hb_fir_filter(float *out, const float fir_coef[HB_FIR_SIZE + 1], + float mem[HB_FIR_SIZE], float cp_gain, const float *in) +{ + int i, j; + float data[AMRWB_SFR_SIZE_OUT + HB_FIR_SIZE]; // past and current samples + + memcpy(data, mem, HB_FIR_SIZE * sizeof(float)); + + for (i = 0; i < AMRWB_SFR_SIZE_OUT; i++) + data[i + HB_FIR_SIZE] = in[i] / cp_gain; + + for (i = 0; i < AMRWB_SFR_SIZE_OUT; i++) + { + out[i] = 0.0; + for (j = 0; j <= HB_FIR_SIZE; j++) + out[i] += data[i + j] * fir_coef[j]; + } + + memcpy(mem, data + AMRWB_SFR_SIZE_OUT, HB_FIR_SIZE * sizeof(float)); +} + +/** * Update context state before the next subframe */ static void update_sub_state(AMRWBContext *ctx) @@ -1212,6 +1247,7 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, 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_SFR_SIZE_OUT]; // excitation for the high frequency band + float hb_samples[AMRWB_SFR_SIZE_OUT]; // filtered high-band samples from synthesis float hb_gain; int sub, i; @@ -1324,12 +1360,17 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, hb_exc, ctx->isf_cur, ctx->isf_past_final); /* High-band post-processing filters */ + hb_fir_filter(hb_samples, bpf_6_7_coef, ctx->bpf_6_7_mem, + 4.0, &ctx->samples_hb[LP_ORDER_16k]); + + if (ctx->fr_cur_mode == MODE_23k85) + hb_fir_filter(hb_samples, lpf_7_coef, ctx->lpf_7_mem, + 1.0, hb_samples); /* Add low frequency and high frequency bands */ for (i = 0; i < AMRWB_SFR_SIZE_OUT; i++) { // XXX: the lower band should really be upscaled by 2.0? - sub_buf[i] = (sub_buf[i] * 2.0 + - ctx->samples_hb[i + LP_ORDER_16k]) / 32768.0; + sub_buf[i] = (sub_buf[i] * 2.0 + hb_samples[i]) / 32768.0; } /* Update buffers and history */ commit 8cbd44b44a1b8d4332fc8d22ced69a93d85a31fc Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Fri Jul 30 02:32:08 2010 -0300 Fix high-band excitation length diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 7f6152f..2786816 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -931,11 +931,10 @@ static void de_emphasis(float *out, float *in, 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 hpf_coef [in] filter coefficients as used below - * @param mem [in] state from last filtering + * @param mem [in/out] state from last filtering (updated) * @param in [in] input speech data * * @remark It is safe to pass the same array in in and out parameters. @@ -1028,13 +1027,13 @@ static void scaled_hb_excitation(AMRWBContext *ctx, float *hb_exc, 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++) + for (i = 0; i < AMRWB_SFR_SIZE_OUT; i++) hb_exc[i] = 32768.0 - (uint16_t) av_lfg_get(&ctx->prng); ff_scale_vector_to_given_sum_of_squares(hb_exc, hb_exc, energy, - AMRWB_SUBFRAME_SIZE); + AMRWB_SFR_SIZE_OUT); - for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) + for (i = 0; i < AMRWB_SFR_SIZE_OUT; i++) hb_exc[i] *= hb_gain; } @@ -1212,7 +1211,7 @@ 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 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_exc[AMRWB_SFR_SIZE_OUT]; // excitation for the high frequency band float hb_gain; int sub, i; ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdata.h | 31 ++++++++++++++++++++++++ libavcodec/amrwbdec.c | 60 ++++++++++++++++++++++++++++++++++++++++-------- 2 files changed, 81 insertions(+), 10 deletions(-) hooks/post-receive -- AMR-WB decoder

3 3

[soc]: r5875 - in mms: mms.h mmsh.c
by spyfeng 29 Jul '10

29 Jul '10

Author: spyfeng Date: Wed Jul 28 17:42:24 2010 New Revision: 5875 Log: make mmsh.c use mms.h Modified: mms/mms.h mms/mmsh.c Modified: mms/mms.h ============================================================================== --- mms/mms.h Wed Jul 28 17:09:11 2010 (r5874) +++ mms/mms.h Wed Jul 28 17:42:24 2010 (r5875) @@ -34,7 +34,7 @@ typedef struct { /** Buffer for outgoing packets. */ /*@{*/ uint8_t *write_out_ptr; ///< Pointer for writting the buffer. - uint8_t out_buffer[512]; ///< Buffer for outgoing packet. + uint8_t out_buffer[8192]; ///< Buffer for outgoing packet. /*@}*/ /** Buffer for incoming packets. */ @@ -55,27 +55,4 @@ typedef struct { int stream_num; ///< stream numbers. } 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 Wed Jul 28 17:09:11 2010 (r5874) +++ mms/mmsh.c Wed Jul 28 17:42:24 2010 (r5875) @@ -68,18 +68,31 @@ static const char* mmsh_live_request = "Pragma: stream-switch-entry=%s\r\n" "Connection: Close\r\n\r\n"; +typedef struct +{ + MMSContext *ff_ctx; + char location[1024]; + int seekable; + int stream_num; + int request_seq; + int chunk_seq; +}MMSHContext; + static int mmsh_close(URLContext *h) { - MMSHContext *mms = (MMSHContext *)h->priv_data; + MMSHContext *mmsh_ctx = (MMSHContext *)h->priv_data; + MMSContext *mms = mmsh_ctx->ff_ctx; if(mms->mms_hd) url_close(mms->mms_hd); - av_freep(&h->priv_data); av_freep(&mms->asf_header); + av_freep(&mms); + av_freep(&h->priv_data); return 0; } -static int asf_header_parser(MMSHContext *mms) +static int asf_header_parser(MMSHContext *mmsh_ctx) { + MMSContext *mms = mmsh_ctx->ff_ctx; uint8_t *p = mms->asf_header; uint8_t *end; int flags, stream_id; @@ -139,8 +152,9 @@ static int asf_header_parser(MMSHContext return 0; } -static int get_chunk_header(MMSHContext *mms, int *len) +static int get_chunk_header(MMSHContext *mmsh_ctx, int *len) { + MMSContext *mms = mmsh_ctx->ff_ctx; uint8_t chunk_header[CHUNK_HEADER_LENGTH]; uint8_t ext_header[EXT_HEADER_LENGTH]; int chunk_type; @@ -170,12 +184,13 @@ static int get_chunk_header(MMSHContext } *len = chunk_len - ext_header_len; if (chunk_type == CHUNK_TYPE_END || chunk_type == CHUNK_TYPE_DATA) - mms->chunk_seq = AV_RL32(ext_header); + mmsh_ctx->chunk_seq = AV_RL32(ext_header); return chunk_type; } -static int read_data_packet(MMSHContext *mms, const int len) +static int read_data_packet(MMSHContext *mmsh_ctx, const int len) { + MMSContext *mms = mmsh_ctx->ff_ctx; int res, pad_size = 0; res = url_read_complete(mms->mms_hd, mms->in_buffer, len); dprintf(NULL, "data packet len = %d\n", len); @@ -191,23 +206,24 @@ static int read_data_packet(MMSHContext memset(mms->in_buffer + len, 0, pad_size); } mms->read_in_ptr = mms->in_buffer; - mms->asf_data_remaining_len = mms->asf_packet_len; + mms->remaining_in_len = mms->asf_packet_len; return 0; } -static int get_http_header_data(MMSHContext *mms) +static int get_http_header_data(MMSHContext *mmsh_ctx) { + MMSContext *mms = mmsh_ctx->ff_ctx; int res, len; int chunk_type; for (;;) { len = 0; - chunk_type = get_chunk_header(mms, &len); + chunk_type = get_chunk_header(mmsh_ctx, &len); if (chunk_type < 0) { return chunk_type; } else if (chunk_type == CHUNK_TYPE_ASF_HEADER){ // get asf header and stored it - if (!mms->is_header_parsed) { + if (!mms->header_parsed) { if (mms->asf_header) { if (len != mms->asf_header_size) { mms->asf_header_size = len; @@ -227,14 +243,14 @@ static int get_http_header_data(MMSHCont return -1; } mms->asf_header_size = len; - if (!mms->is_header_parsed) { - res = asf_header_parser(mms); - mms->is_header_parsed = 1; + if (!mms->header_parsed) { + res = asf_header_parser(mmsh_ctx); + mms->header_parsed = 1; return res; } } else if (chunk_type == CHUNK_TYPE_DATA) { // read data packet and do padding - return read_data_packet(mms, len); + return read_data_packet(mmsh_ctx, len); break; } else { if (len) { @@ -252,8 +268,9 @@ static int get_http_header_data(MMSHCont return 0; } -static int mmsh_open_cnx(MMSHContext *mms) +static int mmsh_open_cnx(MMSHContext *mmsh_ctx) { + MMSContext *mms = mmsh_ctx->ff_ctx; int i, port, err, offset = 0; char httpname[256], path[256], host[128]; char stream_selection[10 * MAX_STREAMS]; @@ -264,7 +281,7 @@ static int mmsh_open_cnx(MMSHContext *mm } ff_url_split(NULL, 0, NULL, 0, - host, sizeof(host), &port, path, sizeof(path), mms->location); + host, sizeof(host), &port, path, sizeof(path), mmsh_ctx->location); if(port<0) port = 80; // defaut mmsh protocol port ff_url_join(httpname, sizeof(httpname), "http", NULL, host, port, path); @@ -274,13 +291,13 @@ static int mmsh_open_cnx(MMSHContext *mm } snprintf (headers, sizeof(headers), mmsh_first_request, - host, port, mms->request_seq++); + host, port, mmsh_ctx->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); + err = get_http_header_data(mmsh_ctx); if (err) { dprintf(NULL, "get http header data fialed!\n"); return (err); @@ -300,12 +317,12 @@ static int mmsh_open_cnx(MMSHContext *mm offset += err; } // send paly request - if (mms->seekable) { + if (mmsh_ctx->seekable) { err = snprintf(headers, sizeof(headers), mmsh_seekable_request, - host, port, 0, 0, 0, mms->request_seq++, 0, mms->stream_num, stream_selection); + host, port, 0, 0, 0, mmsh_ctx->request_seq++, 0, mms->stream_num, stream_selection); } else { err = snprintf(headers, sizeof(headers), mmsh_live_request, - host, port, mms->request_seq++, mms->stream_num, stream_selection); + host, port, mmsh_ctx->request_seq++, mms->stream_num, stream_selection); } if (err < 0) { dprintf(NULL, "build play request failed!\n"); @@ -318,7 +335,7 @@ static int mmsh_open_cnx(MMSHContext *mm return AVERROR(EIO); } - err = get_http_header_data(mms); + err = get_http_header_data(mmsh_ctx); if (err) { dprintf(NULL, "get http header data fialed!\n"); return (err); @@ -329,15 +346,19 @@ static int mmsh_open_cnx(MMSHContext *mm static int mmsh_open(URLContext *h, const char *uri, int flags) { - MMSHContext *mms; + MMSHContext *mmsh_ctx; + MMSContext *mms; int err; - mms = h->priv_data = av_mallocz(sizeof(MMSHContext)); + mmsh_ctx = h->priv_data = av_mallocz(sizeof(MMSHContext)); if (!h->priv_data) return AVERROR(ENOMEM); - mms->request_seq = h->is_streamed = 1; - av_strlcpy(mms->location, uri, sizeof(mms->location)); + mms = mmsh_ctx->ff_ctx = av_mallocz(sizeof(MMSContext)); + if (!mmsh_ctx->ff_ctx) + return AVERROR(ENOMEM); + mmsh_ctx->request_seq = h->is_streamed = 1; + av_strlcpy(mmsh_ctx->location, uri, sizeof(mmsh_ctx->location)); - err =mmsh_open_cnx(mms); + err =mmsh_open_cnx(mmsh_ctx); if (err) { dprintf(NULL, "Leaving mmsh open (failure: %d)\n", err); mmsh_close(h); @@ -347,41 +368,43 @@ static int mmsh_open(URLContext *h, cons return 0; } -static int read_data(MMSHContext *mms, char *buf, int size) +static int read_data(MMSHContext *mmsh_ctx, char *buf, int size) { + MMSContext *mms = mmsh_ctx->ff_ctx; int read_size; - read_size = FFMIN(size, mms->asf_data_remaining_len); + read_size = FFMIN(size, mms->remaining_in_len); memcpy(buf, mms->read_in_ptr, read_size); - mms->asf_data_remaining_len -= read_size; + mms->remaining_in_len -= read_size; mms->read_in_ptr += read_size; return read_size; } -static int handle_chunk_type(MMSHContext *mms) +static int handle_chunk_type(MMSHContext *mmsh_ctx) { + MMSContext *mms = mmsh_ctx->ff_ctx; int res, len = 0; int chunk_type; - chunk_type = get_chunk_header(mms, &len); + chunk_type = get_chunk_header(mmsh_ctx, &len); if(chunk_type == CHUNK_TYPE_END) { - if (mms->chunk_seq == 0) { + if (mmsh_ctx->chunk_seq == 0) { dprintf(NULL, "The stream is end.\n"); return -1; } // reconnect - mms->request_seq = 1; - if ((res = mmsh_open_cnx(mms)) !=0) { + mmsh_ctx->request_seq = 1; + if ((res = mmsh_open_cnx(mmsh_ctx)) !=0) { dprintf(NULL, "Reconnect failed!\n"); return res; } } else if (chunk_type == CHUNK_TYPE_STREAM_CHANGE) { - mms->is_header_parsed = 0; - if ((res = get_http_header_data(mms)) !=0) { + mms->header_parsed = 0; + if ((res = get_http_header_data(mmsh_ctx)) !=0) { dprintf(NULL,"stream changed! get new header failed!\n"); return res; } } else if (chunk_type == CHUNK_TYPE_DATA) { - return read_data_packet(mms, len); + return read_data_packet(mmsh_ctx, len); } else { dprintf(NULL, "recv other type packet %d\n", chunk_type); return -1; @@ -392,8 +415,8 @@ static int handle_chunk_type(MMSHContext static int mmsh_read(URLContext *h, uint8_t *buf, int size) { int res = 0; - MMSHContext *mms = h->priv_data; - + MMSHContext *mmsh_ctx = h->priv_data; + MMSContext *mms = mmsh_ctx->ff_ctx; do{ if (mms->asf_header_read_size < mms->asf_header_size) { // copy asf header into buffer @@ -408,13 +431,13 @@ static int mmsh_read(URLContext *h, uint if (mms->asf_header_read_size == mms->asf_header_size) { av_freep(&mms->asf_header); // which contains asf header } - } else if (mms->asf_data_remaining_len){ - res =read_data(mms, buf, size); + } else if (mms->remaining_in_len){ + res =read_data(mmsh_ctx, buf, size); } else { // read data packet from network - res = handle_chunk_type(mms); + res = handle_chunk_type(mmsh_ctx); if (res == 0) { - res = read_data(mms, buf, size); + res = read_data(mmsh_ctx, buf, size); } else { dprintf(NULL, "other situation!\n"); }

3 2

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 29 Jul '10

29 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 01100290fe3b220e4fbbe6ec16f7f85652d25092 (commit) from 1c399e4d298648f1ad8a70f50507c41fa7b21a62 (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 01100290fe3b220e4fbbe6ec16f7f85652d25092 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Thu Jul 29 17:41:24 2010 -0300 Write the high-band synthesis process for all modes diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index ce5fb63..76a1a60 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -1621,6 +1621,9 @@ static const int16_t isf_init[LP_ORDER] = { 9216, 10240, 11264, 12288, 13312, 14336, 15360, 3840 }; +/* ISF/ISP interpolation coefficients for each subframe */ +static const float isfp_inter[4] = { 0.45, 0.8, 0.96, 1.0 }; + /* Coefficients for FIR interpolation of excitation vector * at pitch lag resulting the adaptive codebook vector */ static const float ac_inter[65] = { diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 6833f23..7f6152f 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -74,6 +74,7 @@ typedef struct { float samples_az[LP_ORDER + AMRWB_SUBFRAME_SIZE]; ///< lower band samples from synthesis at 12.8kHz float samples_up[UPS_MEM_SIZE + AMRWB_SUBFRAME_SIZE]; ///< lower band samples processed for upsampling at 12.8kHz + float samples_hb[LP_ORDER_16k + AMRWB_SFR_SIZE_OUT]; ///< higher band samples from synthesis at 16kHz float hpf_31_mem[4], hpf_400_mem[4]; ///< previous values in the high-pass filters float demph_mem[1]; ///< previous value in the de-emphasis filter @@ -298,17 +299,14 @@ static void isf_set_min_dist(float *isf, float min_spacing, int size) { */ static void interpolate_isp(double isp_q[4][LP_ORDER], const double *isp4_past) { - int i; + int i, k; // 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]; + for (k = 0; k < 3; k++) { + float c = isfp_inter[k]; + for (i = 0; i < LP_ORDER; i++) + isp_q[k][i] = (1.0 - c) * isp4_past[i] + c * isp_q[3][i]; + } } /** @@ -880,7 +878,7 @@ static void pitch_enhancer(float *fixed_vector, float voice_fac) * @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 samples [in/out] pointer to the output speech samples */ static void synthesis(AMRWBContext *ctx, float *lpc, float *excitation, float fixed_gain, const float *fixed_vector, @@ -985,8 +983,6 @@ static void upsample_5_4(float *out, const float *in, int o_size) } else out[i] = ff_dot_productf(in0 + int_part, upsample_fir[4 - frac_part], UPS_MEM_SIZE); - - out[i] *= 2.0 / 32768.0; // upscale output } } @@ -1064,7 +1060,7 @@ static float auto_correlation(float *diff_isf, float mean, int lag) * @param out [out] buffer for extrapolated isf * @param isf [in] input isf vector */ -static void extrapolate_isf(float *out, float *isf) +static void extrapolate_isf(float out[LP_ORDER_16k], float isf[LP_ORDER]) { float diff_isf[LP_ORDER - 2], diff_mean; float *diff_hi = diff_isf - LP_ORDER + 1; // diff array for extrapolated indices @@ -1124,6 +1120,63 @@ static void extrapolate_isf(float *out, float *isf) } /** + * Spectral expand the LP coefficients using the equation: + * y[i] = x[i] * (gamma ** i) + * + * @param out [out] output buffer (may use input array) + * @param lpc [in] LP coefficients array + * @param gamma [in] weighting factor + * @param size [in] LP array size + */ +static void lpc_weighting(float *out, const float *lpc, float gamma, int size) +{ + int i; + float fac = 1.0; + + for (i = 0; i < size; i++) { + out[i] = lpc[i] * fac; + fac *= gamma; + } +} + +/** + * Conduct 20th order linear predictive coding synthesis for the high + * frequency band excitation at 16kHz + * + * @param ctx [in] the context + * @param subframe [in] current subframe index (0 to 3) + * @param samples [in/out] pointer to the output speech samples + * @param exc [in] generated white-noise scaled excitation + * @param isf [in] current frame isf vector + * @param isf_past [in] past frame final isf vector + */ +static void hb_synthesis(AMRWBContext *ctx, int subframe, float *samples, + const float *exc, const float *isf, const float *isf_past) +{ + float hb_lpc[LP_ORDER_16k + 1]; + enum Mode mode = ctx->fr_cur_mode; + + if (mode == MODE_6k60) { + float e_isf[LP_ORDER_16k]; // ISF vector for extrapolation + double e_isp[LP_ORDER_16k]; + + ff_weighted_vector_sumf(e_isf, isf_past, isf, isfp_inter[subframe], + 1.0 - isfp_inter[subframe], LP_ORDER); + + extrapolate_isf(e_isf, e_isf); + isf2isp(e_isf, e_isp); + isp2lp(e_isp, hb_lpc, LP_ORDER_16k / 2); + + lpc_weighting(hb_lpc, hb_lpc, 0.9, LP_ORDER_16k + 1); + } else { + lpc_weighting(hb_lpc, ctx->lp_coef[subframe], 0.6, LP_ORDER + 1); + } + + ff_celp_lp_synthesis_filterf(samples, hb_lpc + 1, exc, AMRWB_SFR_SIZE_OUT, + (mode == MODE_6k60) ? LP_ORDER_16k : LP_ORDER); +} + +/** * Update context state before the next subframe */ static void update_sub_state(AMRWBContext *ctx) @@ -1138,6 +1191,8 @@ static void update_sub_state(AMRWBContext *ctx) LP_ORDER * sizeof(float)); memmove(&ctx->samples_up[0], &ctx->samples_up[AMRWB_SUBFRAME_SIZE], UPS_MEM_SIZE * sizeof(float)); + memmove(&ctx->samples_hb[0], &ctx->samples_hb[AMRWB_SUBFRAME_SIZE], + LP_ORDER_16k * sizeof(float)); memset(ctx->fixed_vector, 0, AMRWB_SUBFRAME_SIZE * sizeof(float)); } @@ -1193,10 +1248,11 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, 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); + isp2lp(ctx->isp[sub], ctx->lp_coef[sub], LP_ORDER / 2); for (sub = 0; sub < 4; sub++) { const AMRWBSubFrame *cur_subframe = &cf->subframe[sub]; + float *sub_buf = buf_out + sub * AMRWB_SFR_SIZE_OUT; /* Decode adaptive codebook */ decode_pitch_vector(ctx, cur_subframe, sub); @@ -1238,14 +1294,14 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, 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); synthesis(ctx, ctx->lp_coef[sub], synth_exc, synth_fixed_gain, synth_fixed_vector, &ctx->samples_az[LP_ORDER]); + /* XXX: Tested against the ref code until here, it "succeeds" at least + * for cases in which the "opencore bug" don't interfere */ + /* Synthesis speech post-processing */ de_emphasis(&ctx->samples_up[UPS_MEM_SIZE], &ctx->samples_az[LP_ORDER], PREEMPH_FAC, ctx->demph_mem); @@ -1253,8 +1309,8 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, high_pass_filter(&ctx->samples_up[UPS_MEM_SIZE], hpf_31_coef, ctx->hpf_31_mem, &ctx->samples_up[UPS_MEM_SIZE]); - upsample_5_4(buf_out + sub * AMRWB_SFR_SIZE_OUT, - &ctx->samples_up[UPS_FIR_SIZE], AMRWB_SFR_SIZE_OUT); + upsample_5_4(sub_buf, &ctx->samples_up[UPS_FIR_SIZE], + AMRWB_SFR_SIZE_OUT); /* High frequency band generation */ high_pass_filter(&ctx->samples_up[UPS_MEM_SIZE], hpf_400_coef, @@ -1265,6 +1321,18 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, scaled_hb_excitation(ctx, hb_exc, synth_exc, hb_gain); + hb_synthesis(ctx, sub, &ctx->samples_hb[LP_ORDER_16k], + hb_exc, ctx->isf_cur, ctx->isf_past_final); + + /* High-band post-processing filters */ + + /* Add low frequency and high frequency bands */ + for (i = 0; i < AMRWB_SFR_SIZE_OUT; i++) { + // XXX: the lower band should really be upscaled by 2.0? + sub_buf[i] = (sub_buf[i] * 2.0 + + ctx->samples_hb[i + LP_ORDER_16k]) / 32768.0; + } + /* Update buffers and history */ update_sub_state(ctx); } ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdata.h | 3 + libavcodec/amrwbdec.c | 106 +++++++++++++++++++++++++++++++++++++++--------- 2 files changed, 90 insertions(+), 19 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 29 Jul '10

29 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 1c399e4d298648f1ad8a70f50507c41fa7b21a62 (commit) from 42461e43223b4da99bbfa00a08e844fa8d6acd85 (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 1c399e4d298648f1ad8a70f50507c41fa7b21a62 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Thu Jul 29 02:51:15 2010 -0300 Finish writing the ISF vector extrapolation (6k60 HB) Fix white noise excitation randomness diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index acfb443..ce5fb63 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -28,8 +28,7 @@ #define LP_ORDER_16k 20 ///< lpc filter order at 16kHz #define UPS_FIR_SIZE 12 ///< upsampling filter size #define UPS_MEM_SIZE 2 * UPS_FIR_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) @@ -39,7 +38,6 @@ #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 AMRWB_P_DELAY_MAX 231 ///< maximum pitch delay value #define AMRWB_P_DELAY_MIN 34 diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 83303ed..6833f23 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -1033,7 +1033,7 @@ static void scaled_hb_excitation(AMRWBContext *ctx, float *hb_exc, /* 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; + hb_exc[i] = 32768.0 - (uint16_t) av_lfg_get(&ctx->prng); ff_scale_vector_to_given_sum_of_squares(hb_exc, hb_exc, energy, AMRWB_SUBFRAME_SIZE); @@ -1042,6 +1042,9 @@ static void scaled_hb_excitation(AMRWBContext *ctx, float *hb_exc, hb_exc[i] *= hb_gain; } +/** + * Calculate auto-correlation for the ISF difference vector + */ static float auto_correlation(float *diff_isf, float mean, int lag) { int i; @@ -1054,10 +1057,19 @@ static float auto_correlation(float *diff_isf, float mean, int lag) return sum; } +/** + * Extrapolate a ISF vector to the 16kHz range (20th order LP) + * used at mode 6k60 LP filter for the high-freq band + * + * @param out [out] buffer for extrapolated isf + * @param isf [in] input isf vector + */ static void extrapolate_isf(float *out, float *isf) { float diff_isf[LP_ORDER - 2], diff_mean; + float *diff_hi = diff_isf - LP_ORDER + 1; // diff array for extrapolated indices float corr_lag[3]; + float est, scale; int i, i_max_corr; memcpy(out, isf, LP_ORDER - 1); @@ -1083,7 +1095,32 @@ static void extrapolate_isf(float *out, float *isf) 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; + + /* Calculate an estimate for ISF(18) and scale ISF based on the error */ + est = 79.65 + (out[2] - out[3] - out[4]) / 6.0; + scale = (FFMIN(est, 76.0) - out[LP_ORDER - 2]) / + (out[LP_ORDER_16k - 2] - out[LP_ORDER - 2]); + // XXX: should divide numerator by 2.0? + + for (i = LP_ORDER - 1; i < LP_ORDER_16k - 1; i++) + diff_hi[i] = scale * (out[i] - out[i - 1]); + + /* Stability insurance */ + for (i = LP_ORDER; i < LP_ORDER_16k - 1; i++) + if (diff_hi[i] + diff_hi[i - 1] < 5.0) { + if (diff_hi[i] > diff_hi[i - 1]) { + diff_hi[i - 1] = 5.0 - diff_hi[i]; + } else + diff_hi[i] = 5.0 - diff_hi[i - 1]; + } + + for (i = LP_ORDER - 1; i < LP_ORDER_16k - 1; i++) + out[i] = out[i - 1] + diff_hi[i]; + + /* XXX: Don't know why this 26214 coefficient, maybe it is not Q8 */ + /* Scale the ISF vector for 16000 Hz */ + for (i = 0; i < LP_ORDER_16k - 1; i++) + out[i] *= 26214 / (float) (1 << 8); } /** ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdata.h | 4 +--- libavcodec/amrwbdec.c | 41 +++++++++++++++++++++++++++++++++++++++-- 2 files changed, 40 insertions(+), 5 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0

[soc]AMR-WB decoder branch, master, updated.
by Marcelo Póvoa 28 Jul '10

28 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 42461e43223b4da99bbfa00a08e844fa8d6acd85 (commit) via 0b3e35bc80eb44f58e222244e165a8edc89e031a (commit) via 33c6d2a788ee89d1340cecf89470ba1446db794c (commit) via 6bcfe4400dde8897b6095f076d0e71119aa14b09 (commit) from 5116d7f3e4f7d3dfb7436506d9d9771e3c0cf2a4 (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 42461e43223b4da99bbfa00a08e844fa8d6acd85 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Thu Jul 29 00:09:30 2010 -0300 Bugfixes: Correct the fixed vector update and past excitation generation diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 3bd6e13..83303ed 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -427,15 +427,15 @@ static void decode_pitch_vector(AMRWBContext *ctx, &ctx->base_pitch_lag, subframe); ctx->pitch_lag_int = pitch_lag_int + (pitch_lag_frac < 0 ? -1 : 0); - pitch_lag_int += pitch_lag_frac > 0; + pitch_lag_int = ctx->pitch_lag_int + (pitch_lag_frac ? 1 : 0); /* 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); + pitch_lag_frac + (pitch_lag_frac > 0 ? 0 : 4), + LP_ORDER, AMRWB_SUBFRAME_SIZE + 1); /* Check which pitch signal path should be used. * 6k60 and 8k85 modes have the ltp flag set to 0 */ @@ -444,6 +444,7 @@ static void decode_pitch_vector(AMRWBContext *ctx, } else { for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) ctx->pitch_vector[i] = 0.18 * exc[i - 1] + 0.64 * exc[i] + 0.18 * exc[i + 1]; + memcpy(exc, ctx->pitch_vector, AMRWB_SUBFRAME_SIZE * sizeof(float)); } } @@ -1100,6 +1101,8 @@ static void update_sub_state(AMRWBContext *ctx) LP_ORDER * sizeof(float)); memmove(&ctx->samples_up[0], &ctx->samples_up[AMRWB_SUBFRAME_SIZE], UPS_MEM_SIZE * sizeof(float)); + + memset(ctx->fixed_vector, 0, AMRWB_SUBFRAME_SIZE * sizeof(float)); } static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, @@ -1226,8 +1229,6 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, 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); update_sub_state(ctx); } commit 0b3e35bc80eb44f58e222244e165a8edc89e031a Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Thu Jul 29 00:05:15 2010 -0300 Correct number of past fixed and pitch gains and modify their orderings for simplicity diff --git a/libavcodec/amrwbdec.c b/libavcodec/amrwbdec.c index 35f74af..3bd6e13 100644 --- a/libavcodec/amrwbdec.c +++ b/libavcodec/amrwbdec.c @@ -63,8 +63,8 @@ typedef struct { 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 pitch_gain[6]; ///< quantified pitch gains for the current and previous five subframes + float fixed_gain[2]; ///< quantified fixed gains for the current and previous subframes float tilt_coef; ///< {beta_1} related to the voicing of the previous subframe @@ -729,22 +729,22 @@ static float *anti_sparseness(AMRWBContext *ctx, if (ctx->fr_cur_mode > MODE_8k85) // no filtering in high modes return fixed_vector; - if (ctx->pitch_gain[4] < 0.6) { + if (ctx->pitch_gain[0] < 0.6) { ir_filter_nr = 0; // strong filtering - } else if (ctx->pitch_gain[4] < 0.9) { + } else if (ctx->pitch_gain[0] < 0.9) { ir_filter_nr = 1; // medium filtering } else ir_filter_nr = 2; // no filtering // detect 'onset' - if (ctx->fixed_gain[4] > 3.0 * ctx->fixed_gain[3]) { + if (ctx->fixed_gain[0] > 3.0 * ctx->fixed_gain[1]) { if (ir_filter_nr < 2) ir_filter_nr++; } else { int i, count = 0; - for (i = 0; i < 5; i++) + for (i = 0; i < 6; i++) if (ctx->pitch_gain[i] < 0.6) count++; if (count > 2) @@ -886,17 +886,17 @@ static void synthesis(AMRWBContext *ctx, float *lpc, float *excitation, float *samples) { ff_weighted_vector_sumf(excitation, ctx->pitch_vector, fixed_vector, - ctx->pitch_gain[4], fixed_gain, AMRWB_SUBFRAME_SIZE); + ctx->pitch_gain[0], fixed_gain, AMRWB_SUBFRAME_SIZE); // emphasize pitch vector contribution in low bitrate modes - if (ctx->pitch_gain[4] > 0.5 && ctx->fr_cur_mode <= MODE_8k85) { + if (ctx->pitch_gain[0] > 0.5 && ctx->fr_cur_mode <= MODE_8k85) { int i; float energy = ff_dot_productf(excitation, excitation, AMRWB_SUBFRAME_SIZE); // XXX: Weird part in both ref code and spec. A unknown parameter // {beta} seems to be identical to the current pitch gain - float pitch_factor = 0.25 * ctx->pitch_gain[4] * ctx->pitch_gain[4]; + float pitch_factor = 0.25 * ctx->pitch_gain[0] * ctx->pitch_gain[0]; for (i = 0; i < AMRWB_SUBFRAME_SIZE; i++) excitation[i] += pitch_factor * ctx->pitch_vector[i]; @@ -1093,8 +1093,8 @@ 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)); + memmove(&ctx->pitch_gain[1], &ctx->pitch_gain[0], 5 * sizeof(float)); + memmove(&ctx->fixed_gain[1], &ctx->fixed_gain[0], 1 * sizeof(float)); memmove(&ctx->samples_az[0], &ctx->samples_az[AMRWB_SUBFRAME_SIZE], LP_ORDER * sizeof(float)); @@ -1165,11 +1165,11 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, cur_subframe->pul_il, ctx->fr_cur_mode); decode_gains(cur_subframe->vq_gain, ctx->fr_cur_mode, - &fixed_gain_factor, &ctx->pitch_gain[4]); + &fixed_gain_factor, &ctx->pitch_gain[0]); pitch_sharpening(ctx, &fixed_sparse, ctx->fixed_vector); - ctx->fixed_gain[4] = + ctx->fixed_gain[0] = ff_amr_set_fixed_gain(fixed_gain_factor, ff_dot_productf(ctx->fixed_vector, ctx->fixed_vector, AMRWB_SUBFRAME_SIZE)/AMRWB_SUBFRAME_SIZE, @@ -1177,22 +1177,22 @@ static int amrwb_decode_frame(AVCodecContext *avctx, void *data, int *data_size, ENERGY_MEAN, energy_pred_fac); /* Calculate voice factor and store tilt for next subframe */ - voice_fac = voice_factor(ctx->pitch_vector, ctx->pitch_gain[4], - ctx->fixed_vector, ctx->fixed_gain[4]); + voice_fac = voice_factor(ctx->pitch_vector, ctx->pitch_gain[0], + ctx->fixed_vector, ctx->fixed_gain[0]); 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]; + ctx->excitation[i] *= ctx->pitch_gain[0]; // XXX: Did not used ff_set_fixed_vector like AMR-NB in order // to retain pitch sharpening done to the fixed_vector - ctx->excitation[i] += ctx->fixed_gain[4] * ctx->fixed_vector[i]; + ctx->excitation[i] += ctx->fixed_gain[0] * ctx->fixed_vector[i]; // XXX: Should remove fractional part of excitation like NB? // I did not found a reference of this in the ref decoder } /* Post-processing of excitation elements */ - synth_fixed_gain = noise_enhancer(ctx->fixed_gain[4], &ctx->prev_tr_gain, + synth_fixed_gain = noise_enhancer(ctx->fixed_gain[0], &ctx->prev_tr_gain, voice_fac, stab_fac); synth_fixed_vector = anti_sparseness(ctx, ctx->fixed_vector, commit 33c6d2a788ee89d1340cecf89470ba1446db794c Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Wed Jul 28 23:18:35 2010 -0300 Improve FIR interpolation table for better precision diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 0966e9c..acfb443 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -1626,23 +1626,23 @@ static const int16_t isf_init[LP_ORDER] = { /* Coefficients for FIR interpolation of excitation vector * at pitch lag resulting the adaptive codebook vector */ static const float ac_inter[65] = { - 0.940000, - 0.856390, 0.632268, 0.337560, 0.059072, - -0.131059, -0.199393, -0.158569, -0.056359, - 0.047606, 0.106749, 0.103705, 0.052062, - -0.015182, -0.063705, -0.073660, -0.046497, - -0.000983, 0.038227, 0.053143, 0.040059, - 0.009308, -0.021674, -0.037767, -0.033186, - -0.013028, 0.010702, 0.025901, 0.026318, - 0.013821, -0.003645, -0.016813, -0.019855, - -0.012766, -0.000530, 0.010080, 0.014122, - 0.010657, 0.002594, -0.005363, -0.009344, - -0.008101, -0.003182, 0.002330, 0.005635, - 0.005562, 0.002844, -0.000627, -0.002993, - -0.003362, -0.002044, -0.000116, 0.001315, - 0.001692, 0.001151, 0.000259, -0.000417, - -0.000618, -0.000434, -0.000133, 0.000063, - 0.000098, 0.000048, 0.000007, 0.000000 + 9.400024e-01, + 8.563843e-01, 6.322632e-01, 3.375854e-01, 5.908203e-02, + -1.310425e-01, -1.994019e-01, -1.585693e-01, -5.633545e-02, + 4.760742e-02, 1.067505e-01, 1.036987e-01, 5.206299e-02, + -1.519775e-02, -6.372070e-02, -7.366943e-02, -4.650879e-02, + -9.765625e-04, 3.820801e-02, 5.316162e-02, 4.003906e-02, + 9.338379e-03, -2.166748e-02, -3.778076e-02, -3.320312e-02, + -1.300049e-02, 1.068115e-02, 2.587891e-02, 2.630615e-02, + 1.379395e-02, -3.662109e-03, -1.678467e-02, -1.983643e-02, + -1.275635e-02, -5.493164e-04, 1.007080e-02, 1.409912e-02, + 1.068115e-02, 2.624512e-03, -5.371094e-03, -9.338379e-03, + -8.117676e-03, -3.173828e-03, 2.319336e-03, 5.615234e-03, + 5.554199e-03, 2.868652e-03, -6.103516e-04, -2.990723e-03, + -3.356934e-03, -2.014160e-03, -1.220703e-04, 1.342773e-03, + 1.708984e-03, 1.159668e-03, 2.441406e-04, -4.272461e-04, + -6.103516e-04, -4.272461e-04, -1.220703e-04, 6.103516e-05, + 1.220703e-04, 6.103516e-05, 0.000000e+00, 0.000000e+00 }; /* [i][j] is the number of pulses present in track j at mode i */ commit 6bcfe4400dde8897b6095f076d0e71119aa14b09 Author: Marcelo Povoa <marspeoplester(a)gmail.com> Date: Wed Jul 28 23:17:45 2010 -0300 Fix typos in the reordering tables for 12k65 mode diff --git a/libavcodec/amrwbdata.h b/libavcodec/amrwbdata.h index 8db1617..0966e9c 100644 --- a/libavcodec/amrwbdata.h +++ b/libavcodec/amrwbdata.h @@ -175,40 +175,40 @@ static const uint16_t order_MODE_12k65[] = { 9, AMR_OF(0, pul_il[3]), 112, 149, 181, 213, 245, 153, 185, 217, 249, 7, AMR_OF(0, vq_gain), 3, 20, 42, 28, 32, 38, 24, - 6, AMR_OF(2, adap), 36, 49, 72, 77, 83, 98, - 1, AMR_OF(2, ltp), 106, - 9, AMR_OF(2, pul_il[0]), 113, 126, 158, 190, 222, 130, 162, 194, + 6, AMR_OF(1, adap), 36, 49, 72, 77, 83, 98, + 1, AMR_OF(1, ltp), 106, + 9, AMR_OF(1, pul_il[0]), 113, 126, 158, 190, 222, 130, 162, 194, 226, - 9, AMR_OF(2, pul_il[1]), 114, 134, 166, 198, 230, 138, 170, 202, + 9, AMR_OF(1, pul_il[1]), 114, 134, 166, 198, 230, 138, 170, 202, 234, - 9, AMR_OF(2, pul_il[2]), 115, 142, 174, 206, 238, 146, 178, 210, + 9, AMR_OF(1, pul_il[2]), 115, 142, 174, 206, 238, 146, 178, 210, 242, - 9, AMR_OF(2, pul_il[3]), 116, 150, 182, 214, 246, 154, 186, 218, + 9, AMR_OF(1, pul_il[3]), 116, 150, 182, 214, 246, 154, 186, 218, 250, - 7, AMR_OF(2, vq_gain), 4, 21, 43, 29, 33, 39, 25, - 9, AMR_OF(3, adap), 15, 16, 17, 18, 19, 51, 70, 80, + 7, AMR_OF(1, vq_gain), 4, 21, 43, 29, 33, 39, 25, + 9, AMR_OF(2, adap), 15, 16, 17, 18, 19, 51, 70, 80, 92, - 1, AMR_OF(3, ltp), 107, - 9, AMR_OF(3, pul_il[0]), 117, 127, 159, 191, 223, 131, 163, 195, + 1, AMR_OF(2, ltp), 107, + 9, AMR_OF(2, pul_il[0]), 117, 127, 159, 191, 223, 131, 163, 195, 227, - 9, AMR_OF(3, pul_il[1]), 118, 135, 167, 199, 231, 139, 171, 203, + 9, AMR_OF(2, pul_il[1]), 118, 135, 167, 199, 231, 139, 171, 203, 235, - 9, AMR_OF(3, pul_il[2]), 119, 143, 175, 207, 239, 147, 179, 211, + 9, AMR_OF(2, pul_il[2]), 119, 143, 175, 207, 239, 147, 179, 211, 243, - 9, AMR_OF(3, pul_il[3]), 120, 151, 183, 215, 247, 155, 187, 219, + 9, AMR_OF(2, pul_il[3]), 120, 151, 183, 215, 247, 155, 187, 219, 251, - 7, AMR_OF(3, vq_gain), 5, 22, 44, 30, 34, 40, 26, - 6, AMR_OF(4, adap), 37, 50, 73, 78, 84, 99, - 1, AMR_OF(4, ltp), 108, - 9, AMR_OF(4, pul_il[0]), 121, 128, 160, 192, 224, 132, 164, 196, + 7, AMR_OF(2, vq_gain), 5, 22, 44, 30, 34, 40, 26, + 6, AMR_OF(3, adap), 37, 50, 73, 78, 84, 99, + 1, AMR_OF(3, ltp), 108, + 9, AMR_OF(3, pul_il[0]), 121, 128, 160, 192, 224, 132, 164, 196, 228, - 9, AMR_OF(4, pul_il[1]), 122, 136, 168, 200, 232, 140, 172, 204, + 9, AMR_OF(3, pul_il[1]), 122, 136, 168, 200, 232, 140, 172, 204, 236, - 9, AMR_OF(4, pul_il[2]), 123, 144, 176, 208, 240, 148, 180, 212, + 9, AMR_OF(3, pul_il[2]), 123, 144, 176, 208, 240, 148, 180, 212, 244, - 9, AMR_OF(4, pul_il[3]), 124, 152, 184, 216, 248, 156, 188, 220, + 9, AMR_OF(3, pul_il[3]), 124, 152, 184, 216, 248, 156, 188, 220, 252, - 7, AMR_OF(4, vq_gain), 6, 23, 45, 31, 35, 41, 27, + 7, AMR_OF(3, vq_gain), 6, 23, 45, 31, 35, 41, 27, 0 }; ----------------------------------------------------------------------- Summary of changes: libavcodec/amrwbdata.h | 76 ++++++++++++++++++++++++------------------------ libavcodec/amrwbdec.c | 47 +++++++++++++++-------------- 2 files changed, 62 insertions(+), 61 deletions(-) hooks/post-receive -- AMR-WB decoder

1 0