Unrecognized mp3 file

[parro-it]

I have some mp3 file not recognized by file-type:

❯ file /usr/local/Wowza/content/20151027_Consiglio_01.mp3
/usr/local/Wowza/content/20151027_Consiglio_01.mp3: MPEG ADTS, layer III,  v2.5,  64 kbps, 11.025 kHz, Monaural

first bytes in file are 0xffe380, while file-type check for 0x494433 or else 0xfffb.

https://en.wikipedia.org/wiki/List_of_file_signatures only list the two checks actually done by file-type, but the magic header file used by posix file command seems to list a more complex check:

# MP3, M1A
0       beshort&0xFFFE  0xFFFA         MPEG ADTS, layer III, v1
# rates
>2      byte&0xF0       0x10           \b,  32 kBits
>2      byte&0xF0       0x20           \b,  40 kBits
>2      byte&0xF0       0x30           \b,  48 kBits
>2      byte&0xF0       0x40           \b,  56 kBits
>2      byte&0xF0       0x50           \b,  64 kBits
>2      byte&0xF0       0x60           \b,  80 kBits
>2      byte&0xF0       0x70           \b,  96 kBits
>2      byte&0xF0       0x80           \b, 112 kBits
>2      byte&0xF0       0x90           \b, 128 kBits
>2      byte&0xF0       0xA0           \b, 160 kBits
>2      byte&0xF0       0xB0           \b, 192 kBits
>2      byte&0xF0       0xC0           \b, 224 kBits
>2      byte&0xF0       0xD0           \b, 256 kBits
>2      byte&0xF0       0xE0           \b, 320 kBits
# timing
>2      byte&0x0C       0x00           \b, 44.1 kHz
>2      byte&0x0C       0x04           \b, 48 kHz
>2      byte&0x0C       0x08           \b, 32 kHz
# channels/options
>3      byte&0xC0       0x00           \b, Stereo
>3      byte&0xC0       0x40           \b, JntStereo
>3      byte&0xC0       0x80           \b, 2x Monaural
>3      byte&0xC0       0xC0           \b, Monaural
#>1     byte            ^0x01          \b, Data Verify
#>2     byte            &0x02          \b, Packet Pad
#>2     byte            &0x01          \b, Custom Flag
#>3     byte            &0x08          \b, Copyrighted
#>3     byte            &0x04          \b, Original Source
#>3     byte&0x03       1              \b, NR: 50/15 ms
#>3     byte&0x03       3              \b, NR: CCIT J.17

[mifi]

Actually libmagic has a lot of checks for identifying different mp3's:
https://github.com/threatstack/libmagic/blob/master/magic/Magdir/animation

Your file would match this rule:

# MP3, M25A
0       beshort&0xFFFE  0xFFE2         MPEG ADTS, layer III,  v2.5
!:mime	audio/mpeg

We could implement the logic from libmagic. However I don't know how we should map to file extension, because MP1, MP2 and MP3 all just resolve to audio/mpeg

[parro-it]

We could implement the logic from libmagic.

I see your call for libmagic reimplementation in js, I would be glad to help with the effort... I think
about 3 possibility to start:

1. try to compile libmagic to js using emscripten

2. do a native binding module

3. reimplement all the stuff

4. or 2) would probably be the better choice, but 3) is more fun for sure 😄

However I don't know how we should map to file extension, because MP1, MP2 and MP3 all just resolve to audio/mpeg

I've no idea... for my use case I need only the mime type. magic header comment seems to speak only
about MP3?

[mifi]

I was actually thinking about just implementing the mp3 stuff from libmagic, not the whole thing.

We made an issue about implementing everything (#68). And I actually tried to use a couple hours to implement the instruction set from libmagic. I got some of them right, but I realized that they have a quite big instruction set, some of which are not trivial to implement, so I gave up. Also it got quite slow.

But the logic required for parsing audio/mpeg is quite trivial to implement, PRs are welcome :)

[sindresorhus]

Also libmagic covers a lot of obscure and ancient stuff we no longer care about. Would be very bloaty to compile it with emscripten. I think a better approach would be to find a way to precompile a selection of the rules into some kind of generated JS.

[parro-it]

I think a better approach would be to find a way to precompile a selection of the rules into some kind of generated JS.

The format of the rules is well explained in the man page, so this should be possible. Only, as @midif said, there are lot of stuff to implement... But it could be done incrementally at least

[s-ol]

I also have a mono mp3 file with 0xFFFA magic. Here's a PR for it in another project: h2non/filetype#11

[karlhiramoto]

In PR #101 It should fix this case as well as other cases I encountered.