TorchAudio constantly fails due to no backend

[PaulLockett]

agents/livekit-plugins/livekit-plugins-openai/livekit/plugins/openai/tts.py

Line 46 in 27ea3c1

tensor, sample_rate = torchaudio.load(io.BytesIO(data), format="mp3")

The synthesize function does not work because Torchaudio has no available backends. I'm trying to find a workaround.

[PaulLockett]

example failure:
ERROR:root:google speech task failed: Couldn't find appropriate backend to handle uri <_io.BytesIO object at 0x28dd8da90> and format mp3.

[PaulLockett]

More context. even with a backend ( that must be in the OS) TorchAudio is not optimized to take in bytes see here: pytorch/audio#800

[keepingitneil]

torchaudio uses installed libraries on the host to do decoding. If you install ffmpeg on your host it should be resolved.

[PaulLockett]

On an m1 Mac this is not so trivia it seems like torch audio does not look at the root directory nor the homebrew directory when searching for dylib files.

[keepingitneil]

torchaudio has been replaced with pyav in this PR: #182