FFmpeg: Decoding of MP3 files fails

[uklotzde]

Bug Description

Caused by #11879.

warning [AnalyzerThread 0 #1] SoundSourceFFmpeg - av_seek_frame() failed: Operation not permitted
warning [AnalyzerThread 0 #1] AudioSource - Failed to read sample frames: expected = [0 -> 1) , actual = [0 -> 0)
info [AnalyzerThread 0 #1] AudioSource - Shrinking readable frame index range: before = [0 -> 16816943) , after = [0 -> 0)
warning [AnalyzerThread 0 #1] AudioSource - Read test failed: expected = [0 -> 1) , actual = [0 -> 0)
warning [AnalyzerThread 0 #1] SoundSourceProxy - Failed to read file "file:///home/uk/Music/1st/*****.mp3" with provider "FFmpeg"

av_seek_frame() fails if seekTimestamp is negative.

Version

No response

OS

No response

[Swiftb0y]

can confirm. #12034

[acolombier]

I am currently experiencing something similar as well. I have been for some times and thought files were missing, but it turns out the are here and can be read just fine with any other software. The is the output of ffprobe on one if them in case it helps:

Input #0, mp3, from './path/to/file':
  Metadata:
    title           : ...
    artist          : ...
    genre           : ...
    TKEY            : 4m
    TBPM            : 128
    id3v2_priv.TRAKTOR4: DMRT\x92\x0d\x00\x00\x02\x00\x00\x00RDH 0\x00\x00\x00\x03\x00\x00\x00SKHC\x04\x00\x00\x00\x00\x00\x00\x00\x81z\x04\x00DOMF\x04\x00\x00\x00\x00\x00\x00\x00\x19\x04\xdf\x07NSRV\x04\x00\x00\x00\x00\x00\x00\x00\x07\x00\x00\x00ATADJ\x0d\x00\x00\x12\x00\x00\x00
  Duration: 00:05:39.39, start: 0.000000, bitrate: 320 kb/s
  Stream #0:0: Audio: mp3, 44100 Hz, stereo, fltp, 320 kb/s

Log output:

info [CachingReaderWorker 1] SoundSourceMp3 - Recoverable MP3 header decoding error: lost synchronization
info [CachingReaderWorker 1] SoundSourceMp3 - MP3 frame header | layer: 0 mode: 0 #channels: 1 #samples: 36 bitrate: 0 samplerate: 0 flags: "0x0000"
info [CachingReaderWorker 1] SoundSourceMp3 - Recoverable MP3 header decoding error: reserved header layer value
[...repeat...]
warning [CachingReaderWorker 1] SoundSourceMp3 - Differing number of channels 2 <> 1 in MP3 frame headers: "/path/to/file"
info [CachingReaderWorker 1] SoundSourceMp3 - Recoverable MP3 frame decoding error: forbidden bit allocation value
warning [CachingReaderWorker 1] SoundSourceMp3 - MP3 frame header with mismatching number of channels 1 <> 2  - aborting
warning [CachingReaderWorker 1] AudioSource - Failed to read sample frames: expected = [0 -> 1) , actual = [0 -> 0)
info [CachingReaderWorker 1] AudioSource - Shrinking readable frame index range: before = [0 -> 14946432) , after = [0 -> 0)
warning [CachingReaderWorker 1] AudioSource - Read test failed: expected = [0 -> 1) , actual = [0 -> 0)
warning [CachingReaderWorker 1] SoundSourceProxy - Failed to read file "file:///path/to/file" with provider "MAD: MPEG Audio Decoder 0.15.1 (beta) NDEBUG FPM_64BIT"
[mp3 @ 0x7fffac00a480] Skipping 1035 bytes of junk at 29444.
[mp3 @ 0x7fffac00a480] Estimating duration from bitrate, this may be inaccurate
warning [CachingReaderWorker 1] SoundSourceFFmpeg - av_seek_frame() failed: Operation not permitted
warning [CachingReaderWorker 1] AudioSource - Failed to read sample frames: expected = [0 -> 1) , actual = [0 -> 0)
info [CachingReaderWorker 1] AudioSource - Shrinking readable frame index range: before = [0 -> 14966926) , after = [0 -> 0)
warning [CachingReaderWorker 1] AudioSource - Read test failed: expected = [0 -> 1) , actual = [0 -> 0)
warning [CachingReaderWorker 1] SoundSourceProxy - Failed to read file "file:///path/to/file" with provider "FFmpeg 4.4.2-0ubuntu0.22.04.1+esm1"
info [CachingReaderWorker 1] SoundSourceMp3 - Recoverable MP3 header decoding error: lost synchronization
info [CachingReaderWorker 1] SoundSourceMp3 - MP3 frame header | layer: 0 mode: 0 #channels: 1 #samples: 36 bitrate: 0 samplerate: 0 flags: "0x0000"
info [CachingReaderWorker 1] SoundSourceMp3 - Recoverable MP3 header decoding error: reserved header layer value
[...repeat...]
warning [CachingReaderWorker 1] SoundSourceMp3 - Differing number of channels 2 <> 1 in MP3 frame headers: "/path/to/file"
info [CachingReaderWorker 1] SoundSourceMp3 - Recoverable MP3 frame decoding error: forbidden bit allocation value
warning [CachingReaderWorker 1] SoundSourceMp3 - MP3 frame header with mismatching number of channels 1 <> 2  - aborting
warning [CachingReaderWorker 1] AudioSource - Failed to read sample frames: expected = [0 -> 1) , actual = [0 -> 0)
info [CachingReaderWorker 1] AudioSource - Shrinking readable frame index range: before = [0 -> 14946432) , after = [0 -> 0)
warning [CachingReaderWorker 1] AudioSource - Read test failed: expected = [0 -> 1) , actual = [0 -> 0)
warning [CachingReaderWorker 1] SoundSourceProxy - Failed to read file "file:///path/to/file" with provider "MAD: MPEG Audio Decoder 0.15.1 (beta) NDEBUG FPM_64BIT"
[mp3 @ 0x7fffac00a480] Skipping 1035 bytes of junk at 29444.
[mp3 @ 0x7fffac00a480] Estimating duration from bitrate, this may be inaccurate
warning [CachingReaderWorker 1] SoundSourceFFmpeg - av_seek_frame() failed: Operation not permitted
warning [CachingReaderWorker 1] AudioSource - Failed to read sample frames: expected = [0 -> 1) , actual = [0 -> 0)
info [CachingReaderWorker 1] AudioSource - Shrinking readable frame index range: before = [0 -> 14966926) , after = [0 -> 0)
warning [CachingReaderWorker 1] AudioSource - Read test failed: expected = [0 -> 1) , actual = [0 -> 0)
warning [CachingReaderWorker 1] SoundSourceProxy - Failed to read file "file:///path/to/file" with provider "FFmpeg 4.4.2-0ubuntu0.22.04.1+esm1"
warning [CachingReaderWorker 1] SoundSourceProxy - Giving up to open file "file:///path/to/file" after 4 unsuccessful attempts
warning [CachingReaderWorker 1] CachingReaderWorker - "[Channel1]" Failed to open file QFileInfo(/path/to/file)

Is this still an issue and related or would that be something different? The open PR seems to only include a test change.

Edit: the SoundSourceFFmpeg - av_seek_frame() failed: Operation not permitted makes me think it is probably the same

[Swiftb0y]

Well, can you try #12034 and see if that fixes it for you too. Then the issue is pretty clear cut...

[acolombier]

The open PR seems to only include a test change.

Or do you mean something suggested in the PR discussion?

[Swiftb0y]

Ah I forgot that the actual fix got rebased out halfway through the PR. If you clamp the seeking (see 012a415) then you should be able to load the mp3s. However, as pointed out by the others in that PR, it reintroduces some issues with playing back AACs so thats why I didn't get accepted. I have not bothered enough to look into this, though we really should.

[uklotzde]

I have reverted the "fix" locally and use FFmpeg for MP3/M4A. No issues.

No one seems to be able to come up with a suitable offset compensation or otherwise prove that the usage of the FFmpeg API in Mixxx is wrong.

Everyone seems to be happy with the broken state...

[acolombier]

Thanks for the tip @Swiftb0y, will give it a go asap.

No one seems to be able to come up with a suitable offset compensation or otherwise prove that the usage of the FFmpeg API in Mixxx is wrong.

VLC/ffplay seem to be able to deal fine with those mp3s, and I would hope AACs too (I don't use that format so can't verify), perhaps it's worth to check the implementation in there and see if there is any indication on something we are doing wrong. I will try to give that a go.

[uklotzde]

VLC/ffplay seem to be able to deal fine with those mp3s, and I would hope AACs too (I don't use that format so can't verify), perhaps it's worth to check the implementation in there and see if there is any indication on something we are doing wrong.

The offset compensation for MP3 that has been added later as a "fix" is wrong. There were never any issues with decoding. VLC/ffplay don't need to care about offsets when decoding lossy formats.

[daschuer]

The original issue we see here is that the SoundSourceMp3 is not able to decode the file. Than SoundSourceFFmpeg fails as well probably because of this bug. You may send the effected file to my email and I can have a look why our error recovering code fails in your case.

[acolombier]

I've just shared it to your Mixxx address via Drive. I can also send it to you directly if you would prefer (not sure if drive perform any post treatment after upload on MP3)

[daschuer]

Received it and I can confirm the issue.

[acolombier]

I haven't been experienced this issue since your fix @daschuer
Perhaps we can close this issue?

[daschuer]

The issue is still kind of open because it is an issue with the not standard configuration of not using libmad for mp3.
ffmpeg uses to he faster libmpg123, wich is unfortunately not compatible to the waveforms and cues created with libmad.
So a switch is not trivial.

[uklotzde]

Decoding MP3 with FFmpeg is still broken. I am using a local patch that basically reverts the non-workimg changes.

[Swiftb0y]

same here. Been using ffmpeg since 2.3-beta iirc... The incompatibility boils down to the varying start-timing between the decoders, or is there something else?

[daschuer]

Yes basically. There will also be issues when skipping frames in the process of error handling but I would rather ignore that.