FFmpegSource

[Chainmax]

I want to open a DXA video in Avisynth, do I need anything else than FFMPEGSource?

[Myrsloik]

This Is A Version Only Intended For Testing, Idiots Should Skip This Post

This test version uses a new way to decode audio without using a huge uncompressed audio cache. Matroska files need no additional space and all other containers have the original demuxed audio track stored in c:\ffmpegsource.cache (this test version only).

The decoding results seems to be identical (dithering/other parts of codecs that prevent it but makes no difference) when run multiple times when compared to itself. It does however differ a lot from ffmpegsource(atrack=-1) and I have no idea why that is. There are at least no audible artifacts (to me) so report if it works properly.

Usage:
FFAudioSource(string source)

Link

[Wilbert]

@Myrsloik,

Recently MLP decoding was added: ffmpeg-mailinglist. Could you add this to FFAudioSource? I would be very grateful!

[Myrsloik]

I tried to compile a newer ffmpeg the usual way but I get this error:

gcc -shared -Wl,--output-def,avformat-51.def -Wl,--enable-runtime-pseudo-reloc -Wl,--enable-auto-image-base -Wl,--warn-common -Wl,-rpath-link,"/c/libraries/ffmpeg"/libavcodec -Wl,-rpath-link,"/c/libraries/ffmpeg"/libavformat -Wl,-rpath-link,"/c/libraries/ffmpeg"/libavutil -o avformat-51.dll allformats.o cutils.o os_support.o sdp.o utils.o avio.o aviobuf.o raw.o aiff.o riff.o amr.o apc.o ape.o asf.o asfcrypt.o au.o avidec.o avs.o vocdec.o voc.o bethsoftvid.o c93.o daud.o dsicin.o dv.o dxa.o electronicarts.o ffm.o flic.o flvdec.o 4xm.o gifdec.o gxf.o idcin.o img2.o ipmovie.o matroskadec.o matroska.o mm.o mmf.o mov.o isom.o mp3.o mpc.o mpeg.o mpegts.o mtv.o mxf.o nsvdec.o nutdec.o nut.o nuv.o ogg2.o oggparsevorbis.o oggparsetheora.o oggparseflac.o oggparseogm.o rtsp.o rmdec.o idroq.o segafilm.o siff.o smacker.o sol.o psxstr.o swf.o thp.o tiertexseq.o tta.o txd.o sierravmd.o wav.o wc3movie.o westwood.o wv.o yuv4mpeg.o file.o http.o rtpproto.o tcp.o udp.o -L"/c/libraries/ffmpeg"/libavutil -lavutil -lavcodec -L"/c/libraries/ffmpeg"/libavcodec -lz -lm -la52 -lfaad  -lws2_32 
c:/libraries/ffmpeg/libavcodec/libavcodec.a(cscd.o): In function `decode_frame':
c:/libraries/ffmpeg/libavcodec/cscd.c:161: undefined reference to `lzo1x_decode'
c:/libraries/ffmpeg/libavcodec/libavcodec.a(nuv.o): In function `decode_frame':
c:/libraries/ffmpeg/libavcodec/nuv.c:168: undefined reference to `lzo1x_decode'
c:/libraries/ffmpeg/libavcodec/libavcodec.a(cook.o): In function `scalar_dequant_float':
c:/libraries/ffmpeg/libavutil/random.h:45: undefined reference to `av_random_generate_untempered_numbers'
c:/libraries/ffmpeg/libavcodec/libavcodec.a(cook.o): In function `cook_decode_init':
c:/libraries/ffmpeg/libavcodec/cook.c:1069: undefined reference to `av_init_random'
c:/libraries/ffmpeg/libavcodec/libavcodec.a(mpc.o): In function `mpc7_decode_frame':
c:/libraries/ffmpeg/libavutil/random.h:45: undefined reference to `av_random_generate_untempered_numbers'
c:/libraries/ffmpeg/libavcodec/libavcodec.a(mpc.o): In function `mpc7_decode_init':
c:/libraries/ffmpeg/libavcodec/mpc.c:83: undefined reference to `av_init_random'
c:/libraries/ffmpeg/libavcodec/libavcodec.a(nellymoserdec.o): In function `decode_init':
c:/libraries/ffmpeg/libavcodec/nellymoserdec.c:401: undefined reference to `av_init_random'
c:/libraries/ffmpeg/libavcodec/libavcodec.a(nellymoserdec.o): In function `nelly_decode_block':
c:/libraries/ffmpeg/libavutil/random.h:45: undefined reference to `av_random_generate_untempered_numbers'
collect2: ld returned 1 exit status
make[1]: *** [avformat-51.dll] Error 1
make[1]: Leaving directory `/c/libraries/ffmpeg/libavformat'
make: *** [lib] Error 2

It seems like it can't find the libraries it's compiled itself or something similar. No idea what to do about it.

[Myrsloik]

New compile of the test version. Includes a recent ffmpeg and it's now statically linked (flac cache not enabled since it didn't want to link).

Link

[Wilbert]

Thx! I got it working for AAC and MP4/AAC, although i didn't test it very well. It seems that the MLP/TrueHD patches are not added to the SVN yet, but i will keep an eye on it.

[tateu]

@Myrsloik,

You might want to try "strip -x libavcodec.a" on all the ffmpeg libs for the static compile. In my tests, that cuts the resulting dll size down to about 3MB.

[Wilbert]

@Myrsloik,

Kurtnoise created a patch for E-AC3/MLP (ffmpeg)! Could you incorporate that in FFAudioSource?

patch: http://kurtnoise.free.fr/index.php?dir=misc/

Thanks again!

[Myrsloik]

If I did you would have no motivation to get it integrated into the main branch. The source is still available if someone really wants to make their own build with an "improved" ffmpeg and I don't mind answering compilation questions... AS LONG AS YOU CLEARLY LABEL IT AS NOT COMPILED BY ME. You could even get your own link in the first post! Apply now!

[Kurtnoise]

@Wilbert: Done... but not tested.

@Myrsloik: why using FLAC as external lib ? FFmpeg has his own FLAC decoder...

[Leak]

Originally Posted by Kurtnoise13

@Myrsloik: why using FLAC as external lib ? FFmpeg has his own FLAC decoder...

I was under the impression he also needed a FLAC encoder for the audio cache - does ffmpeg do FLAC encoding as well?

np: Prefuse-73 - Every Party Has A Winner And A Loser (Unrest)

[Kurtnoise]

Originally Posted by Leak

I was under the impression he also needed a FLAC encoder for the audio cache - does ffmpeg do FLAC encoding as well?

yes it does...

[Wilbert]

@Kurtnoise13,

http://rapidshare.com/files/66801815...mples.zip.html contains some samples: AAC, MP4/AAC and MLP.

The AAC, MP4/AAC played, but the sound was a bit garbled.

The MLP (44.1 kHz, 16 bit) didn't play at all, it gave the error: 'Audio decoding error'. Script:

v = BlankClip()
a = FFmpegSource("D:\Install\dvd-audio\TheCorrs\DVDAExplorer\Acapela1.mlp", vtrack=-2, atrack=-1)
AudioDub(v,a)

http://lists.mplayerhq.hu/pipermail/...er/036639.html contains a raw-mlp support patch. I don't know whether that was included in your patch? (Perhaps related, perhaps not )

[Myrsloik]

The reason is very simple, I'm lazy.

Actually libflac has a nice api that allows sample accurate seeking all the time and is extremely simple to use. With libavformat seeking is much less fun and where you end up is usually a bit uncertain (if it wasn't, someone else probably would've made ffmpegsource before me).

[Kurtnoise]

Originally Posted by Wilbert

I don't know whether that was included in your patch? (Perhaps related, perhaps not )

Yes, you're right. I forgot this. New build uploaded. Should be fine now...

About aac multichannel files, dunno why it doesn't work properly actually...I'll try something else later.

[Wilbert]

Yes, you're right. I forgot this. New build uploaded. Should be fine now...

It works, i'm so excited!!!!!!!! Thanks you very much!

One additional request though 24 bit stuff is reported as 16 by AviSynth. I'm not sure whether it is converted to 16 bit down the road, or if the patch doesn't support 24 bit. I got an example here:

http://rapidshare.com/files/67023394...ples2.zip.html

Perhaps you can look at it. If the patch doesn't support it, i will contact Ian Caulfield.

[Myrsloik]

From avcodec.h:

/* Currently unused, may be used if 24/32 bits samples are ever supported. */
/* all in native-endian format */
enum SampleFormat {
    SAMPLE_FMT_NONE = -1,
    SAMPLE_FMT_U8,              ///< unsigned 8 bits
    SAMPLE_FMT_S16,             ///< signed 16 bits
    SAMPLE_FMT_S24,             ///< signed 24 bits
    SAMPLE_FMT_S32,             ///< signed 32 bits
    SAMPLE_FMT_FLT,             ///< float
};

Further down...

    /**
     * audio sample format
     * - encoding: Set by user.
     * - decoding: Set by libavcodec.
     */
    enum SampleFormat sample_fmt;  ///< sample format, currently unused

[Wilbert]

That sounds bad. Does that mean that outputting 24 bit is still not supported in ffmpeg?

[tebasuna51]

Originally Posted by Wilbert

That sounds bad. Does that mean that
outputting 24 bit is still not supported in ffmpeg?

Yep, ffmpeg convert anything to 16 bit and preserve the original channel mapping (of course the ac3/aac multichannel conversions are wrong channel mapped).

I think we need preserve the original precision (bitdepth) and use a standard channel mapping for decoder-encoder compatibility.

Wilbert, don't know this thread?
Justin answer about ffmeg bitdepth and multichannel behaviour, there are also a solution to work with NeroAacEnc stdin requested by you in other thread.

[Kurtnoise]

From mlp_parser.c :

#ifdef CONFIG_AUDIO_NONSHORT
        avctx->bits_per_sample = mh.group1_bits;
        if (avctx->bits_per_sample > 16)
            avctx->sample_fmt = SAMPLE_FMT_S32;
#endif