[FFmpeg-devel] AMR-WB+ support in Ffmpeg

[Benoit Fouet]

Hi,

On 12/15/2008 08:07 AM, Igor Gueths wrote:
> Hi all. I was attempting to add AMR-WB+ support into my working copy of R-16140;
> however, my modifications did not work as expected, and I wasn't sure if I had
> missed anything along the way. I am as yet not too familiar with the current
> Ffmpeg API, which may have something to do with it. Up to now, I have done the
> following:
> I downloaded the reference code for TS-26.304 from
> http://www.3gpp.org/ftp/specs/archive/26_series/26.304/26304-700.zip. After
> extracting the file, I successfully built a shared library (I called it
> libamrwb.so.1.0.0 for testing purposes
>  used to build the library was located in the c-code/lib_amr directory. After
> building the library, I started diffing the AMR-WB+ code against the original
> AMR-WB floating point codec, by starting with dec_main.c. I found that a lot of
> the function calls and variable declarations were identical, so I went ahead
> with trying to link Ffmpeg against the new library. Before I could attempt to
> recompile Ffmpeg though, I found that configure complained about Libamr_WB not
> being found. In looking through the undefined symbols, I found that these were
> gain_hf_ramp, oversamp_12k8, smooth_ener_hf, and soft_exc_hf. I found references
> to these in dec_util.c, and found that they had only been prototyped, but had
> been called in the file anyway; therefore, removing the prototypes and
> function calls, rebuilding the library, and copying it back into place along
> with the headers fixed the problem (configure was able to compile a test program
> this time around).

AMR support in FFmpeg is done through libamr [1] which is included in
most distros today (libamrnb-dev and libamrwb-dev), that's why configure
checks for their existences. If you build it manually, you have to make
it possible to locate it.
Also, AFAIK, libamr does not support WB+ at the moment...

>  Before running make though, I had taken a statement out of
> libavcodec/libamr.c, as shown in this diff.
> libamr.c.orig       2008-12-14 02:17:26.948370518 -0500
> +++ libamr.c    2008-12-14 02:42:31.849369919 -0500
> @@ -643,12 +643,6 @@
>
>      amr_decode_fix_avctx(avctx);
>
> -    if(avctx->channels > 1)
> -    {
>   

just changing the test should be enough, no need to remove it completely

> -        av_log(avctx, AV_LOG_ERROR, "amr_wb: multichannel decoding not
>          supported\n");
> -        return -1;
> -    }
> -
>      return 0;
>  }
>
>
> Make ran ok; I got the Ffmpeg binary reinstalled, and everything looked to be
> ok. However, when I finally went to decode my test.3gp file, whose audio is
> encoded in AMR-WB+, I got:
> ffmpeg -acodec libamr_wb -i test.3gp test.wav
>   

can you share this sample (as in uploading it to mplayer ftp) ?

> FFmpeg version SVN-r16140, Copyright (c) 2000-2008 Fabrice Bellard, et al.
>   configuration: --prefix=/usr --enable-gpl --enable-libamr-wb
> --enable-libamr-nb --enable-libfaac --enable-libfaad --enable-libmp3lame
> --enable-libspeex --enable-libvorbis --enable-libx264 --enable-libxvid
> --enable-nonfree --enable-ffmpeg
> --enable-shared --enable-libgsm
>   libavutil     49.12. 0 / 49.12. 0
>   libavcodec    52. 6. 3 / 52. 6. 3
>   libavformat   52.23. 1 / 52.23. 1
>   libavdevice   52. 1. 0 / 52. 1. 0
>   built on Dec 14 2008 20:09:33, gcc: 4.1.2
> Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'test.3gp':
>   Duration: 00:00:02.40, start: 0.000000, bitrate: 21 kb/s
>     Stream #0.0(und): Audio: libamr_wb, 6464 Hz, stereo, s16
>   

the sampling frequency seems weird.

> Output #0, wav, to 'test.wav':
>     Stream #0.0(und): Audio: pcm_s16le, 6464 Hz, stereo, s16, 206 kb/s
> Stream mapping:
>   Stream #0.0 -> #0.0
> Press [q] to stop encoding
> Multiple frames in a packet from stream 0
> Multiple frames in a packet from stream 0
> Multiple frames in a packet from stream 0
> I keep getting that message indefinitely, and my test.wav output file keeps
> growing with presumably null bytes. 
>
> Did I in fact miss something when attempting to link against my test AMR-WB+
> library? Perhaps there is more that needs to be done in libavcodec/libamr.c to
> get the decoding working properly?

I guess there is more to do, yes (even though I've not investigated)

>  Or, is AMR-WB+ already supported somewhere,
> and I just have not found the correct patch yet? Any pointers or advice that you
> may be able to provide with regards to my above process would be appreciated.
> Thanks!
>   
>   

Ben

[1]: http://www.penguin.cz/~utx/amr