Unable to save audio file

[haveyouwantto]

🐛 Bug Report

Error while saving the audio file.

To Reproduce

Run command demucs out.m4a --two-stems vocals -n htdemucs_ft

Expected behavior

The audio file appears in output folder.

Actual Behavior

The program throws an error and the audio file is truncated.

Your Environment

• Python and PyTorch version: Python 3.9.13 Pytorch 1.12.0+cu116
• Operating system and version (desktop or mobile): Windows 10 22H2
• Hardware (gpu or cpu, amount of RAM etc.): GTX 1660 super 16GB RAM

[CarlGao4]

Please run python -c "import soundfile;print(soundfile.__version__);print(soundfile.__libsndfile_version__)" and paste the outputs here.

[haveyouwantto]

>>> soundfile.__version__
'0.11.0'
>>> soundfile.__libsndfile_version__
'1.1.0'

[CarlGao4]

I think that the soundfile you are using has been modified, because line 316 is f.write(data) in your screenshot, but the original repository is comment subtype : str, optional.
I suggest reinstalling soundfile using the package manager you installed them.

[haveyouwantto]

I reinstalled soundfile from pypi, and now it throws another execption.

Traceback (most recent call last):
  File "D:\Python\lib\runpy.py", line 197, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "D:\Python\lib\runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "D:\Python\Scripts\demucs.exe\__main__.py", line 7, in <module>
  File "D:\Python\lib\site-packages\demucs\separate.py", line 192, in main
    save_audio(sources.pop(model.sources.index(args.stem)), str(stem), **kwargs)
  File "D:\Python\lib\site-packages\demucs\audio.py", line 254, in save_audio
    ta.save(str(path), wav, sample_rate=samplerate,
  File "D:\Python\lib\site-packages\torchaudio\backend\soundfile_backend.py", line 435, in save
    soundfile.write(file=filepath, data=src, samplerate=sample_rate, subtype=subtype, format=format)
  File "D:\Python\lib\site-packages\soundfile.py", line 342, in write
    f.write(data)
  File "D:\Python\lib\site-packages\soundfile.py", line 1017, in write
    written = self._array_io('write', data, len(data))
  File "D:\Python\lib\site-packages\soundfile.py", line 1341, in _array_io
    return self._cdata_io(action, cdata, ctype, frames)
  File "D:\Python\lib\site-packages\soundfile.py", line 1351, in _cdata_io
    _error_check(self._errorcode)
  File "D:\Python\lib\site-packages\soundfile.py", line 1404, in _error_check
    raise LibsndfileError(err, prefix=prefix)
soundfile.LibsndfileError: System error.

[CarlGao4]

Things become difficult. It's an error occured in libsndfile - the core of soundfile, but it can't print out the error. What I can do is just opening an issue under soundfile repository.

[haveyouwantto]

PS: It only happens with demucs v4, the v3 branch is fine.

[CarlGao4]

It's becoming more interesting. I'll do some experience on my computer.

[CarlGao4]

Please help me test one thing:
Add ;print(str(stem), sources[model.sources.index(args.stem)].shape) to the end of the file D:\Python\lib\site-packages\demucs\separate.py line 191
then run your command again
paste the outputs here

[haveyouwantto]

I found out it will only happen when the input audio is long ( > 20 minutes )

[haveyouwantto]

Oh, I found out the problem, my disk is almost full.