[FFmpeg-devel] MPEG Audio elementary streams and layers

[Marc Mason]

Hello,

CODEC_ID_MP2 and CODEC_ID_MP3 are defined in avcodec.h

As far as I understand,
CODEC_ID_MP2 = MPEG Audio Layer II
CODEC_ID_MP3 = MPEG Audio Layer III

CODEC_ID_MP3 appeared in rev 2231 with the following comment.
/* preferred ID for MPEG Audio layer 1, 2 or 3 decoding */
http://svn.ffmpeg.org/ffmpeg/trunk/libavcodec/avcodec.h?r1=2217&r2=2231

What does the comment mean ?


ff_mpegaudio_decode_header() determines the layer of the ES.
http://cekirdek.pardus.org.tr/~ismail/ffmpeg-docs/mpegaudiodecheader_8c-source.html#l00033
s->layer = 4 - ((header >> 17) & 3);

ff_mpa_decode_header() then copies the info to avctx->sub_id
http://cekirdek.pardus.org.tr/~ismail/ffmpeg-docs/mpegaudio__parser_8c-source.html#l00047
avctx->sub_id = s->layer;


When av_find_stream_info() is given an MPEG Audio Layer II elementary
stream, it returns CODEC_ID_MP3 with sub_id=2

This seems ambiguous to me.

What was the original intent ? How is one supposed to determine the
layer of an MPEG Audio elementary stream ?

I can see two possibilities.

1.) Use codec_id

i.e. libavformat returns codec_id = one of CODEC_ID_MP1, CODEC_ID_MP2,
CODEC_ID_MP3 according to the layer.

Then ff_mpa_decode_header() should set the field to CODEC_ID_MP2 when
appropriate.

Something along the lines of

$ svn diff
Index: mpegaudio_parser.c
===================================================================
--- mpegaudio_parser.c  (revision 16054)
+++ mpegaudio_parser.c  (working copy)
@@ -62,6 +62,7 @@
          break;
      case 2:
          avctx->frame_size = 1152;
+        avctx->codec_id = CODEC_ID_MP2;
          break;
      default:
      case 3:

However, what happens with layer I ES ?
CODEC_ID_MP1 might have to be defined ?


2.) Use sub_id

i.e. codec_id indicates MPEG Audio and sub_id indicates the layer of the
MPEG Audio elementary stream.

Then CODEC_ID_MP3 is a misnomer, which really stands for
CODEC_ID_MPEGAUDIO (or CODEC_ID_MPA, or ...) and CODEC_ID_MP2 should be
deleted.


Which is the correct hypothesis ? Neither perhaps ?

I am willing to write a patch either way.

-- 
Regards.