Audioread "no backend" issue

[ghost]

This comes up very frequently for Windows users. pip install -r requirements.txt does not install a backend for audioread (a dependency of librosa). This causes an exception when loading any file that requires conversion, such as .mp3.

Can we make use of soundfile to load and convert audio files, instead of librosa.load? We already have soundfile in requirements.txt and it automatically installs all prerequisites on Windows and macOS. Linux users need to install libsndfile1.

This will prevent users from encountering the "no backend" error. Alternatively we can add a step to README.md asking the user to install ffmpeg before running the toolbox. But this seems more elegant if it works.

[ghost]

Can someone from the community look into this? First, to see if soundfile can be a complete replacement for librosa.load() without the need to install a backend like ffmpeg. If that is the case, then submit a pull request replacing librosa.load() with the corresponding soundfile call in demo_cli.py and encoder/audio.py . Maybe some other places.

[ghost]

Soundfile does not support mp3 (which is a popular input type judging from issues #116, #198, #214, #289, #385, #395, #407) so I think it is best to continue supporting librosa.load() which requires ffmpeg.

The new recommendation is to update README.md to list ffmpeg as a requirement.

[ghost]

This issue will be resolved by PR #414 .