December 2005 - ffmpeg-devel - lists.ffmpeg.org

BUG REPORT: Garbled audio when AC3 stream changes channel count
by nr＠eecs.harvard.edu 28 Jan '06

28 Jan '06

FFmpeg becomes confused when transcoding certain audio files. The resulting audio files are OK up to a point, after which everything sounds like chipmunks. I suspect that the problem is an A/52 (AC3) audio stream that changes between 5.1 channels and 2 channels in the middle of the stream. Details follow. I am using the CVS version of FFmpeg compiled today, 18 Dec 2005. * What I am trying to accomplish: transcode US ATSC high-definition television broadcast (MPEG-2 TS) to MPEG-2 file suitable for passing to dvdauthor and burning to DVD. * The problem I encountered: at some point partway through a file, audio suddenly sounds like 'the chipmunks': it is very high-pitched, unintelligble, and no longer synchronized with video. Audio typically then finishes well before video. * The exact command line I was using: myffmpeg -i bud-light.ts -target ntsc-dvd -b 4000 -bufsize 400 -ab 384 \ -maxrate 7000 -mbd 2 -trell zzz -acodec ac3 ffm-bud-light.mpg N.B. 'myffmpeg' is my version compiled from CVS (as opposed to the Debian version in /usr/bin/ffmpeg). * The console output provided by FFmpeg: ffmpeg version CVS, build 3342336, Copyright (c) 2000-2004 Fabrice Bellard configuration: --enable-gpl --enable-a52 built on Dec 18 2005 12:48:26, gcc: 4.0.3 20051111 (prerelease) (Debian 4.0.2-4) Input #0, mpegts, from 'bud-light.ts': Duration: 00:00:25.3, start: 17010.607378, bitrate: 17551 kb/s Stream #0.0[0x31], 59.94 fps(r): Video: mpeg2video, yuv420p, 1280x720, 16800 kb/s Stream #0.1[0x34](eng): Audio: ac3, 48000 Hz, 5:1, 384 kb/s Output #0, dvd, to 'ffm-bud-light.mpg': Stream #0.0, 29.97 fps(c): Video: mpeg2video (hq), yuv420p, 720x480, q=2-31, 4000 kb/s Stream #0.1: Audio: ac3, 48000 Hz, 5:1, 384 kb/s Stream mapping: Stream #0.0 -> #0.0 Stream #0.1 -> #0.1 Press [q] to stop encoding [mpeg2video @ 0x833d388]Warning MVs not available [mpeg2video @ 0x833d388]concealing 3320 DC, 3320 AC, 3320 MV errors frame= 763 q=0.0 Lsize= 12572kB time=18.0 bitrate=5737.0kbits/s video:11304kB audio:842kB global headers:0kB muxing overhead 3.515292% N.B. Although the clip runs 25 seconds, FFmpeg reports only 18 seconds. This report is consistent with the output file, which contains 25 seconds of video and 18 seconds of audio. It is not useful to run ffmpeg_g, because ffmpeg completes execution without errors---only the output is bad. Sample media illustrating the problem may be found at http://http://www.eecs.harvard.edu/~nr/ffmpeg/chipmunks/bud-light.ts This bug report may be found as file README in that directory. I will be happy to upload this file to ftp.mplayerhq.hu, but even this 25-second clip is 53MB (greater than the recommended 10MB). HD files are large :-( I suspect this bug might be a fairly easy one to fix, if only I knew the code base. Probably the audio decoder is assuming that the format of the AC3 stream remains unchanged. Perhaps the problem could be fixed by borrowing a more pessimistic decoder from the 'a52dec' package. The 'a52dec' application plays the AC3 stream correctly. Norman Ramsey

11 24

[PATCH] from DivX, Part 6: clean types
by slhomme＠divxcorp.com 25 Jan '06

25 Jan '06

Mostly to avoid compiler warnings, and also make it better for 64 bits compilers. The patch in matroska also renames the variable because apparently it didn't like that one. (probably a #define to something else in the system includes)

3 2

Patch DURATION_MAX_READ_SIZE in lavf/utils.c
by rothgang＠uiuc.edu 24 Jan '06

24 Jan '06

The setting of DURATION_MAX_READ_SIZE in lavf/utils.c is a bit too small to get ahead of the last I frame in one of the chapters of "Run Lola Run". Didn't titrate for an exact amount. This patch fixes the problem, and hopefully isn't too excessive. -- Fred

3 5

[PATCH] WinCE support (via gcc cross-compiling)
by gbazin＠altern.org 23 Jan '06

23 Jan '06

Hi, For the WinCE port of VLC media player, we use a modified version of the arm-pe-wince port of gcc + some mingw32 stuff to compile all our stuff. (see https://trac.videolan.org/vlc/file/trunk/INSTALL.wince for the compiling instructions). So here is a patch that adds an --enable-mingwce config option to the build and fixes a few compiling problems with WinCE. Cheers, -- Gildas

6 13

[RFC] Support building universal binaries of FFmpeg on Darwin
by danchr＠opendarwin.org 22 Jan '06

22 Jan '06

Hi, Attached below is an initial patch to allow building a universal, or fat, binary of FFmpeg on Darwin 8.0 and later using GCC 4.0. The generated binary will contain 4 architectures: G3, G4, G5 and x86. Having the three PowerPC architectures allows GCC to tune for the specified CPU, and possibly vectorize code. The configure script needed a lot of hackery to defer detection of e.g. AltiVec to compile-time. Source code changes mainly involve ensuring that files in libavcodec/ppc and libavcodec/i386 compile with both PPC and x86 compilers. A few cases where CONFIG_DARWIN was incorrectly assumed to mean Motorola AltiVec syntax was also fixed. However, I ran into a major issue: The Darwin assembler is an old version of GNU as, and it does not support the i386 syntax used in FFmpeg. I believe the issues boil down to: - The asm macro syntax used in Darwin[1] is different from the currently used syntax. It appears to not support named arguments and use ".endmacro" instead of ".endm". Perhaps using C macros would be better? (In i386/fdct_mmx.c) - The .balign pseudo-op doesn't exist. The best substitute I could find was .align[2]. (In i386/dsputil_mmx.c) - I get a lot of errors about "Junk character 47 (/)". I don't know what that means :/ (In i386/simple_idct_mmx.c) I don't know x86 assembly, so I have no idea how to fix this. I was hoping someone else might be able to help. All other files compile fine. - Dan Christiansen P.S. Please Cc any replies directly to me, as I'm not subscribed to either of these lists. Alex Beregszaszi said on IRC that this might be of interest to MPlayer developers as well, so this mail is Cc'ed there. [1] <http://developer.apple.com/documentation/DeveloperTools/Reference/Assembler…> [2] <http://developer.apple.com/documentation/DeveloperTools/Reference/Assembler…>

2 1

[PATCH] small build cleanup + new configure option
by gbazin＠altern.org 21 Jan '06

21 Jan '06

Hi, Here is a small patch that makes sure --disable-muxers and --disable-ffserver work properly. A new --disable-protocols option is also added to the build to disable the I/O protocols support from libavformat. For instance VLC media player can make use of the libavformat demuxers but uses its own I/O routines, furthermore the libavformat I/O protocols routines don't compile under WinCE so this option can be handy. Anyway, apply if you find this useful. Cheers, -- Gildas

3 15

DV decoding crash
by sms＠2BSD.COM 16 Jan '06

16 Jan '06

Hi - It looks like the massive changes over the last couple days broke DV decoding. In libavcodec/dv.c the 'dv_anchor' array was made part of 'DVVideoContext' - a "per context" (thread?) structure. However, 'dv_anchor' is only initialized in the "global" context. In 'dvvideo_init()' the logic is: static int done=0; if (!done) { s->dv_anchor = av_malloc(12*27*sizeof(void*)); ... } Thus dv_anchor is only initialized once on the first call to dvvideo_init() which would come from calling avcodec_open. After that, further calls to avcodec_open() will NOT initialize dv_anchor leaving it NULL and causing a crash. I'm not sure which is the correct fix - put 'dv_anchor' back to being a static as it was (which worked ;)) OR moving the initialization of dv_anchor outside the 'if (!done)' block so that it gets initialized for each context. It almost makes some sense, seeing how dv_anchor is used, to make it a static common to all threads/contexts. All that dv_anchor holds is a once computed array: for (i=0; i<12*27; i++) s->dv_anchor[i] = (void*)(size_t)i; I'll experiment a bit when I get back from lunch ;) Cheers, Steven Schultz

4 7

[patch] make SONAME encoding optional
by jakemsr＠jakemsr.com 14 Jan '06

14 Jan '06

the SONAME stuff is wrong for OpenBSD. SONAMES should have both major and minor numbers there. but in general, SONAMES aren't used in OpenBSD. patch below adds a --disable-sonames option to disable SONAME encoding. I figure this is more flexible than adding OS specific stuff. the patch also changes the libpostproc Makefile to use LIBVERSION in SHFLAGS like the other library Makefiles do. -- <jakemsr at jakemsr.com> Index: configure =================================================================== RCS file: /cvsroot/ffmpeg/ffmpeg/configure,v retrieving revision 1.220 diff -u -r1.220 configure --- configure 17 Dec 2005 18:14:25 -0000 1.220 +++ configure 18 Dec 2005 01:38:27 -0000 @@ -79,6 +79,7 @@ echo " --enable-small optimize for size instead of speed" echo " --enable-memalign-hack emulate memalign, interferes with memory debuggers" echo " --disable-strip disable stripping of executables and shared libraries" +echo " --disable-sonames disable encoding SONAME into shared libraries" echo " --enable-codec=CODEC enables CODEC" echo " --disable-codec=CODEC disables CODEC" echo " --disable-encoders disables all encoders" @@ -205,6 +206,7 @@ optimize="yes" debug="yes" dostrip="yes" +dosonames="yes" installstrip="-s" extralibs="-lm" simpleidct="yes" @@ -215,7 +217,6 @@ dlfcn="no" dlopen="no" mpegaudio_hp="yes" -SHFLAGS='-shared -Wl,-soname,$@.$(LIBVERSION)' netserver="no" need_inet_aton="no" ffserver="yes" @@ -568,6 +569,8 @@ ;; --disable-strip) dostrip="no" ;; + --disable-sonames) dosonames="no" + ;; --enable-codec=*) CODEC_LIST="$CODEC_LIST ${opt#--enable-codec=}" ;; --disable-codec=*) CODEC_LIST="`echo $CODEC_LIST | sed -e \"s#${opt#--disable-codec=}##\"`" @@ -1244,6 +1247,7 @@ echo "pp support $pp" echo "debug symbols $debug" echo "strip symbols $dostrip" +echo "use SONAME $dosonames" echo "optimize $optimize" echo "shared pp $shared_pp" echo "video hooking $vhook" @@ -1295,6 +1299,13 @@ echo "STRIP=echo ignoring strip" >> config.mak echo "INSTALLSTRIP=" >> config.mak fi +if test -z "$SHFLAGS" ; then + if test "$dosonames" = "yes" ; then + SHFLAGS='-shared -Wl,-soname,$@.$(LIBVERSION)' + else + SHFLAGS="-shared" + fi +fi # SHCFLAGS is a copy of CFLAGS without -mdynamic-no-pic, used when building # shared modules on OS/X (vhook/Makefile). Index: libavcodec/libpostproc/Makefile =================================================================== RCS file: /cvsroot/ffmpeg/ffmpeg/libavcodec/libpostproc/Makefile,v retrieving revision 1.26 diff -u -r1.26 Makefile --- libavcodec/libpostproc/Makefile 26 Sep 2005 10:05:13 -0000 1.26 +++ libavcodec/libpostproc/Makefile 18 Dec 2005 01:38:28 -0000 @@ -12,7 +12,7 @@ else SPPLIBMAJOR = $(SPPLIBNAME).$(SPPMAJOR) SPPLIB = $(SPPLIBNAME).$(SPPVERSION) -SHFLAGS = -shared -Wl,-soname,$(SPPLIB) +LIBVERSION = $(SPPLIB) endif endif PPLIB = $(LIBPREF)postproc$(LIBSUF)

5 15

MJPEG decoding error
by slhomme＠divxcorp.com 13 Jan '06

13 Jan '06

Hi, I have an AVI file with MJPEG that can't be reencoded using FFMPEG, although it plays fine in FFDShow. I verified and the code is similar. The crash in FFMPEG happens when resampling the image before encoding. I think that code is not used in FFDShow. The URL to the crashing sample is : http://www.matroska.org/~robux4/drdivx/wiz011.avi Anyone could investigate this ? I checked the latest CVS under MinGW and I have the crash. thanks PS: for the record, the latest DV patch doesn't work when you call the encoder twice (using libavcodec as a library called many times).

3 3

FAAC rate control broken?
by lucabe72＠email.it 10 Jan '06

10 Jan '06

Hi all, I am trying to encode some AAC audio with ffmpeg (+ libfaac, version 1.24), but it seems that the "-ab" option does not work... Here is a way to reproduce the problem: 1) capture some PCM audio: ./ffmpeg -ac 2 test.wav 2) encode it with aac at 60kbps: ./ffmpeg -i test.wav -ab 64 -acodec aac a.wav ffmpeg version CVS, build 3342336, Copyright (c) 2000-2004 Fabrice Bellard configuration: --cc=gcc-4.0 --enable-faac --disable-network --disable-zlib --disable-vhook --disable-ffplay --extra-cflags=-I /tmp/aac/include --extra-ldflags=-L /tmp/aac/lib built on Dec 21 2005 00:30:58, gcc: 4.0.3 20051201 (prerelease) (Debian 4.0.2-5) Input #0, wav, from 'test.wav': Duration: 00:00:35.5, start: 0.000000, bitrate: 1411 kb/s Stream #0.0: Audio: pcm_s16le, 44100 Hz, stereo, 1411 kb/s File 'a.wav' already exists. Overwrite ? [y/N] y Output #0, wav, to 'a.wav': Stream #0.0: Audio: aac, 44100 Hz, stereo, 64 kb/s Stream mapping: Stream #0.0 -> #0.0 Press [q] to stop encoding size= 356kB time=35.5 bitrate= 82.1kbits/s (so, the output bitrate is 82.1kbps) 3) Encode at 40kbps ./ffmpeg -i test.wav -ab 40 -acodec aac b.wav ffmpeg version CVS, build 3342336, Copyright (c) 2000-2004 Fabrice Bellard configuration: --cc=gcc-4.0 --enable-faac --disable-network --disable-zlib --disable-vhook --disable-ffplay --extra-cflags=-I /tmp/aac/include --extra-ldflags=-L /tmp/aac/lib built on Dec 21 2005 00:30:58, gcc: 4.0.3 20051201 (prerelease) (Debian 4.0.2-5) Input #0, wav, from 'test.wav': Duration: 00:00:35.5, start: 0.000000, bitrate: 1411 kb/s Stream #0.0: Audio: pcm_s16le, 44100 Hz, stereo, 1411 kb/s Output #0, wav, to 'b.wav': Stream #0.0: Audio: aac, 44100 Hz, stereo, 40 kb/s Stream mapping: Stream #0.0 -> #0.0 Press [q] to stop encoding size= 356kB time=35.5 bitrate= 82.1kbits/s (still 82.1kbps...) So, it seems that the bitrate control is not working very well... :( I tried to have a look at libavcodec/faac.c, and it seems that the problem is that Faac_encode_init() does not set faac_cfg->bandWidth... As an experiment, I tried to set it to avctx->bit_rate / avctx->channels / 2 and things look a little bit better: ./ffmpeg -i test.wav -ab 40 -acodec aac a.wav ffmpeg version CVS, build 3342336, Copyright (c) 2000-2004 Fabrice Bellard configuration: --cc=gcc-4.0 --enable-faac --disable-network --disable-zlib --disable-vhook --disable-ffplay --extra-cflags=-I /tmp/aac/include --extra-ldflags=-L /tmp/aac/lib built on Dec 21 2005 00:30:58, gcc: 4.0.3 20051201 (prerelease) (Debian 4.0.2-5) Input #0, wav, from 'test.wav': Duration: 00:00:35.5, start: 0.000000, bitrate: 1411 kb/s Stream #0.0: Audio: pcm_s16le, 44100 Hz, stereo, 1411 kb/s File 'a.wav' already exists. Overwrite ? [y/N] y Output #0, wav, to 'a.wav': Stream #0.0: Audio: aac, 44100 Hz, stereo, 40 kb/s Stream mapping: Stream #0.0 -> #0.0 Press [q] to stop encoding size= 275kB time=35.5 bitrate= 63.5kbits/s still the wrong bitrate (63.5 instead of 40), but it is less than with -ab 64... Does anyone have any idea about how to fix the problem? Thanks, Luca -- _____________________________________________________________________________ Copy this in your signature, if you think it is important: N O W A R ! ! ! -- Email.it, the professional e-mail, gratis per te: http://www.email.it/f Sponsor: Rc auto Zuritel. Risparmiare ? un gioco da ragazzi. Prova ora. Clicca qui: http://adv.email.it/cgi-bin/foclick.cgi?mid=4098&d=20-12

5 10