Valid files(checked by FFprobe) aren't properly parsed in Symphonia

[qarmin]

This files fails to open with latest git symphonia version - 8f4aaed

FFprobe doesn't show any error with them

Code used to parse files - https://github.com/qarmin/audio_checker/blob/master/src/lib.rs

Example files
5 - The Music Of Stasis - Stars And The Moons - Meesah Kuteyi feat Mark Morgan.mp3.zip
08 - Crush.mp3.zip
idw_testsound.wav.zip
InFlux Original Soundtrack - 07 - Volcano Dance.mp3.zip
jetpack1.wav.zip
King of Dragon Pass - 24 - Danger.mp3.zip
Muzyka.ac3.zip
Neverwinter Nights 2 Storm of Zehir soundtrack - maintheme.mp3.zip
Robin Cook - Pandemia (54).mp3.zip
sample.wav.zip
The Way It Was.mp3.zip

[torokati44]

I don't know if their license allows looking at it, or taking inspiration from it, but FFmpeg does a more elaborate check during MP3 frame header synchronization; looking ahead in the buffer and scoring any candidates, then picking the one that looks best: https://github.com/FFmpeg/FFmpeg/blob/38238b604fb59ebaafe93dcf72e544cb33c4ac70/libavformat/mp3dec.c#L499-L547=

[pdeljanov]

I didn't try them all, but for all the MP3 files I tried, the issue seems to be within the ID3v2 parser being too strict. It's encountering a tag with size == 0 which is not allowed by the spec. and returning an error. That error then gets propagated and causes the entire stream to be rejected. With a bit of work, the parser can be made more lenient to allow this case.

The WAV files seems to have a similar issue, but I'd need to look more deeply.

As for the AC3 file, Symphonia does not support AC3. :)

---

I don't think gaining mild inspiration is problematic, but FFMpeg is LGPL licensed and Symphonia is MPL licensed so copying isn't allowed since our license has fewer restrictions. Either way, for these issues it doesn't seem to be the packet parsing that's the issue.

[torokati44]

Either way, for these issues it doesn't seem to be the packet parsing that's the issue.

Okay, that's good! I just thought that it is a regression in 8f4aaed as it was explicitly mentioned. Sorry for the confusion.

[pdeljanov]

The broken MP3s should be fixed with the last two commits.

[pdeljanov]

Regarding the wave files:

• sample.wav and idw_testsound.wav are ADPCM Wave files and are unsupported currently. ADPCM support #41 tracks ADPCM support.
• jetpack1.wav is technically an invalid encoding, but Symphonia should probably handle it.

[qarmin]

More mp3 files with problems
01 Can't Believe My Eyes.mp3.zip
13 Ekspozycja.mp3.zip
55-Steve Jobs Biografia - Rozdzial 26 cz1.mp3.zip
59-Steve Jobs Biografia - Rozdzial 28 cz1.mp3.zip
0086 festung_breslau.mp3.zip
Grendel.mp3.zip

[qarmin]

2023 update

01_Lekcja01.mp3.zip
3_LIST_SW_JANA.mp3.zip
09. Мария Пахоменко - Стоят Девчонки (ill fame Remix).mp3.zip
09 - Goddess.mp3.zip
19-Dzienniki kolymskie.mp3.zip
20-Dzienniki kolymskie.mp3.zip
35 - Moj przyjaciel Kaligula.mp3.zip
59-Steve Jobs Biografia - Rozdzial 28 cz1.mp3.zip
71-Steve Jobs Biografia - Rozdzial 32 cz2.mp3.zip
0086 festung_breslau.mp3.zip
Avril Lavigne - Here's To Never Growing Up.mp3.zip

[sscobici]

See below analysis of the files in 0.6 branch:
Note If "ffmpeg - no errors" is not stated, then there are errors in ffmpeg.
Probably, a separate issue can be opened for point 1. to support mixed files (mono and stereo), as the standard doesn't restrict that. The rest seem to be broken files. All of them are playable except for double speed playback for mono files on Windows, which is expected.d, see #281 (comment)

1. mp3 with different types of frames: starts with mono, continue with stereo - there is a TODO in the code to implement support for this

0086 festung_breslau.mp3
	constain bad MP3 frame at offset: 0x0015202A  (mpa: invalid audio buffer signal spec for packet)
	starts with mono, 64kbs and then switch to stereo, 320kbs
	mono
	
Robin Cook - Pandemia (54).mp3
	constain bad MP3 frame at offset: 0x00099839 ( mpa: invalid audio buffer signal spec for packet )
	starts with mono, 96kbs and then switch to stereo, 320kbs
	plays ok

2. invalid main_data offset, ffmpeg produces no errors (probably weak ffmpeg validation)

3_LIST_SW_JANA.mp3
        fails on first packet at offset: 827842 ( mpa: invalid main_data offset) 
	plays ok
	ffmpeg - no errors

01 Can't Believe My Eyes.mp3
	constain bad MP3 frame at offset: 0x00B4609D, almost at the end of file ( mpa: invalid main_data offset )
	plays ok,
	ffmpeg - no errors

5 - The Music Of Stasis - Stars And The Moons - Meesah Kuteyi feat Mark Morgan.mp3
	constain bad MP3 frame at offset: 0x0039F67E, almost at the end of file ( mpa: invalid main_data offset )
	plays ok,
	ffmpeg - no errors

Grendel.mp3
	constain bad MP3 frame at offset: 0x00A9DFA2 almost at the end of file ( mpa: invalid main_data offset )
	plays ok,
	ffmpeg - no errors

The Way It Was.mp3
	constain bad MP3 frame at offset: 0x003BAB19 almost at the end of file ( mpa: invalid main_data offset )
	plays ok,
	ffmpeg - no errors

3. ID3v1 tag is present in the middle of the file, this is not allowed.

35 - Moj przyjaciel Kaligula.mp3
	contains ID3v1 in the middle of MP3 frames at offset: 0x00048557, Usual frame size is 209. next mp3 frame starts from tag offset + usual frame size
	ID3v1 is not expected in the middle of MP3 frames. 
	mono

55-Steve Jobs Biografia - Rozdzial 26 cz1.mp3
	contains ID3v1 in the middle of MP3 frames at offset: 0x0005A84F, Usual frame size is 208. next mp3 frame starts from tag offset + usual frame size
	ID3v1 is not expected in the middle of MP3 frames. 
	mono

59-Steve Jobs Biografia - Rozdzial 28 cz1.mp3
	contains ID3v1 in the middle of MP3 frames at offset: 0x0003A11F, Usual frame size is 208. next mp3 frame starts from tag offset + usual frame size
	ID3v1 is not expected in the middle of MP3 frames. 
	mono

71-Steve Jobs Biografia - Rozdzial 32 cz2.mp3
	contains ID3v1 in the middle of MP3 frames at offset: 0x00035C0F, Usual frame size is 208. next mp3 frame starts from tag offset + usual frame size
	ID3v1 is not expected in the middle of MP3 frames. 
	mono

19-Dzienniki kolymskie.mp3
	contains ID3v1 in the middle of MP3 frames at offset: 0x00042FD0, Usual frame size is 417. next mp3 frame starts from tag offset + usual frame size
	ID3v1 is not expected in the middle of MP3 frames. 
	mono
	
20-Dzienniki kolymskie.mp3
	contains ID3v1 in the middle of MP3 frames at offset: 0x0008F09B, Usual frame size is 417. next mp3 frame starts from tag offset + usual frame size
	ID3v1 is not expected in the middle of MP3 frames. 
	mono

13 Ekspozycja.mp3
	contains ID3v1 in the middle of MP3 frames at offset: 0x000AB7F2, Usual frame size is 366. next mp3 frame starts from tag offset + usual frame size
	ID3v1 is not expected in the middle of MP3 frames. 
	mono
	
Neverwinter Nights 2 Storm of Zehir soundtrack - maintheme.mp3
	contains broken ID3v1 in the middle of MP3 frames at offset:: 0x00364328, next mp3 frame starts from tag offset + usual frame size
	ID3v1 is not expected in the middle of MP3 frames. 

4. granule big_values > 288, this is not allowed

Avril Lavigne - Here's To Never Growing Up.mp3
	constain bad MP3 frame at offset: 0x000F320B  ( mpa: granule big_values > 288 )
	plays ok

08 - Crush.mp3
	constain bad MP3 frame at offset: 0x0011EC7B ( mpa: granule big_values > 288 )
	plays ok

5. Other

09. Мария Пахоменко - Стоят Девчонки (ill fame Remix).mp3
        error ( mpa: invalid audio buffer signal spec for packet )
	contains ID3v1 at the end at offset: 0x00999EBD but it has only 127 bytes, but should have 128 bytes
	ID3v1 is allowed at the end of the file.
	plays ok

09 - Goddess.mp3
	constain bad MP3 frame at offset: 0x000C5B31 ( mpa: stereo channel pair block_type mismatch )
	plays ok,
	ffmpeg - no errors
	
InFlux Original Soundtrack - 07 - Volcano Dance.mp3
	constain bad MP3 frame at offset: 0x003565CB ( [FAIL] packet=    9045, frame=  10418782 (  47), plane=  0, dec=-0.00000373, ref=+0.00000632 (+0.00001005) )
	a lot of failed packets
	plays ok