mp4 atom - How to discriminate the audio codec? Is it AAC or MP3?

Question

I'm working on a mp4 container parser but I'm going crazy trying to recognize the audio codecs of the streams. I used both QtAtomViewer and AtomicParsley but when I find the atom:

trak->mdia->minf->stbl->stsd

I get always "mp4a" even if the mp4 file has an mp3 stream.

Should I look for an ".mp3" fourcc?

I attach two different mp4 structures: mp4 container with AAC audio stream

     Atom trak @ 716882 of size: 2960, ends @ 719842
     Atom tkhd @ 716890 of size: 92, ends @ 716982
     Atom mdia @ 716982 of size: 2860, ends @ 719842
         Atom mdhd @ 716990 of size: 32, ends @ 717022
         Atom hdlr @ 717022 of size: 33, ends @ 717055
         Atom minf @ 717055 of size: 2787, ends @ 719842
             Atom dinf @ 717063 of size: 36, ends @ 717099
                 Atom dref @ 717071 of size: 28, ends @ 717099
             Atom stbl @ 717099 of size: 2727, ends @ 719826
                 Atom stts @ 717107 of size: 24, ends @ 717131
                 Atom stsz @ 717131 of size: 1268, ends @ 718399
                 Atom stsc @ 718399 of size: 40, ends @ 718439
                 Atom stco @ 718439 of size: 32, ends @ 718471
                 Atom stss @ 718471 of size: 1264, ends @ 719735
                 Atom stsd @ 719735 of size: 91, ends @ 719826
                     Atom mp4a @ 719751 of size: 75, ends @ 719826
                         Atom esds @ 719787 of size: 39, ends @ 719826
             Atom smhd @ 719826 of size: 16, ends @ 719842

mp4 container with mp3 audio stream

Atom trak @ 1663835 of size: 4844, ends @ 1668679
     Atom tkhd @ 1663843 of size: 92, ends @ 1663935
     Atom mdia @ 1663935 of size: 4744, ends @ 1668679
         Atom mdhd @ 1663943 of size: 32, ends @ 1663975
         Atom hdlr @ 1663975 of size: 45, ends @ 1664020
         Atom minf @ 1664020 of size: 4659, ends @ 1668679
             Atom smhd @ 1664028 of size: 16, ends @ 1664044
             Atom dinf @ 1664044 of size: 36, ends @ 1664080
                 Atom dref @ 1664052 of size: 28, ends @ 1664080
             Atom stbl @ 1664080 of size: 4599, ends @ 1668679
                 Atom stsd @ 1664088 of size: 87, ends @ 1664175
                     Atom mp4a @ 1664104 of size: 71, ends @ 1664175
                         Atom esds @ 1664140 of size: 35, ends @ 1664175
                 Atom stts @ 1664175 of size: 24, ends @ 1664199
                 Atom stsc @ 1664199 of size: 28, ends @ 1664227
                 Atom stsz @ 1664227 of size: 2228, ends @ 1666455
                 Atom stco @ 1666455 of size: 2224, ends @ 1668679

Thanks FE

UPDATE:

I found a way to solve the problem: by watching the code of AtomicParsley I see that it's possible to get the codec informations about the stream atom (mp4a), reading the 11th Byte into the esds (Elementary Stream Description) atom.

Now I'm working in this way:

if the value of the 11th Byte is 0x40 I assume the stream is AAC, else if I read 0x69 I assume that the stream is MP3.

I don't like these "empirics" solutions so I'm looking for a more correct way, but I found onlyUnderstanding_AAC that is not complete.

Anyone know where I can get a more detailed specification of MP4 containers?

Answer 1

In the 'esds' atom there are a few fields relevant to determining the codec. The first byte of content of the esds atom is the objectTypeIndication (that's the 11th byte from your solution). This field is supposed to indicate the codec used, but there are a few entries used by multiple codecs. MP4RA has a full list of codec values. Here are few that are relevant in this case:

• 0x40 - MPEG-4 Audio
• 0x6B - MPEG-1 Audio (MPEG-1 Layers 1, 2, and 3)
• 0x69 - MPEG-2 Backward Compatible Audio (MPEG-2 Layers 1, 2, and 3)
• 0x67 - MPEG-2 AAC LC

0x6B and 0x69 denote MPEG-1 and 2 respectively layers 1, 2, and 3. 0x67 denotes MPEG-2 AAC LC but generally is unused in favor of 0x040 (0x66 and 0x68 are also MPEG-2 AAC profiles are seen even less frequently). 0x40 denotes MPEG-4 Audio. MPEG-4 Audio generally is thought of as AAC but there is a whole framework of audio codecs that can go in MPEG-4 Audio including AAC, BSAC, ALS, CELP, and something called MP3On4. MP3On4 is an MP3 variant with some new header information for multichannel.

We can figure out what audio format is actually in the MPEG-4 Audio by looking at the the AudioSpecificConfig. This is the global header for the decoder that exists at byte 13 of the content of the 'esds' atom. At the beginning of the AudioSpecificConfig there is a 5-bit AudioObjectType. A full list can be found on the multimedia wiki (that was linked in your post under the 'MPEG-4 Audio' article: http://wiki.multimedia.cx/index.php?title=MPEG-4_Audio but here are the useful values:

• 00 - NULL
• 01 - AAC Main (a deprecated AAC profile from MPEG-2)
• 02 - AAC LC or backwards compatible HE-AAC (Most realworld AAC falls in one of these cases)
• 03 - AAC Scalable Sample Rate (rarely used)
• 03 - AAC LTP (a replacement for AAC Main, rarely used)
• 05 - HE-AAC explicitly signaled (Non-backward compatible)
• 22 - ER BSAC (A Korean broadcast codec)
• 23 - Low Delay AAC
• 29 - HE-AACv2 explicitly signaled (In one draft this was MP3On4 instead)
• 31 - ESCAPE (read 6 more bits, add 32)
• 32 - MP3on4 Layer 1
• 33 - MP3on4 Layer 2
• 34 - MP3on4 Layer 3

If you aren't worried about 'MP3On4' mp3 variant nor the other weird MPEG-4 Audio codecs then just use the objectTypeIndication.

In the MPEG specifications these details are spread across 14496-1, -12, -14, and -3. Of these only 14496-12 is freely available: http://standards.iso.org/ittf/PubliclyAvailableStandards/index.html