MAC M3 audio backend no longer appearing

[tval2]

🐛 Description

I get "RuntimeError: Couldn't find appropriate backend to handle uri <_io.BytesIO object at 0x13f1f8450> and format None." when processing basic audio data on my mac but not on my colab notebook.

Full code is here `import torchaudio print(torchaudio.list_audio_backends()) def read_audio_from_stream(audio_stream, sampling_rate: int = 16000): sox_backends = set(['sox', 'sox_io']) audio_backends = torchaudio.list_audio_backends()

if len(sox_backends.intersection(audio_backends)) > 0:
    effects = [
        ['channels', '1'],  # Force mono
        ['rate', str(sampling_rate)]  # Resample to the target sampling rate
    ]

    wav, sr = torchaudio.sox_effects.apply_effects_file(audio_stream, effects=effects)
else:
    # Load the audio file from the stream
    wav, sr = torchaudio.load(audio_stream)

    # If more than one channel, take the mean to make it mono
    if wav.size(0) > 1:
        wav = wav.mean(dim=0, keepdim=True)

    # Resample if necessary
    if sr != sampling_rate:
        transform = torchaudio.transforms.Resample(orig_freq=sr, new_freq=sampling_rate)
        wav = transform(wav)
        sr = sampling_rate

assert sr == sampling_rate, "Sampling rate does not match the target sampling rate"
return wav.squeeze(0)  `

where torchaudio.list_audio_backends() always prints to empty []. I tried uninstalling and then reinstalling with `pip install torch torchaudio -f https://download.pytorch.org/whl/torch_stable.html`, I also installed sox. ffmpeg is already installed on my mac. I've seen this pop up on stack overflow 3 times this month as well but no one has come up with an answer. It seems pretty trivial to work for torchaudio.

Versions

torch==2.3.0 torchaudio==2.3.0

[eisneim]

same here, reinstall ffmpeg:

brew install ffmpeg@6 brew install sox

then reinstall torchaudio won't work, torchaudio.list_audio_backends() always prints to empty []

hardware: mac osx m3 max

[JoshuaWilsonBlack]

I've been experiencing the same issue.

I think I have figured out a solution but don't have time to properly test what I've done. I've followed the advice here: https://pytorch.org/audio/stable/build.linux.html in a fresh conda environment, including the "[Optional] Build TorchAudio with a custom built FFmpeg" section. I've just change the line to install ffmpeg to conda install -c conda-forge ffmpeg=6 (i.e., I've specified that the ffmpeg libraries should come from ffmpeg 6.something).

Please forgive this uninformative message!

[eisneim]

here is my solution on mac: brew install ffmpeg@5 and make sure to the lib to path:

export PATH="/opt/homebrew/opt/ffmpeg@6/bin:$PATH"
export LDFLAGS="-L/opt/homebrew/opt/ffmpeg@6/lib"
export CPPFLAGS="-I/opt/homebrew/opt/ffmpeg@6/include"

then reinstall torchaudio