audio

audio copied to clipboard

Hello all!

Currently, TorchAudio doesn’t provide much support for music use cases. We’d like to gauge interest from the community in our improving that support. Some requests we’ve received include operators for chroma feature extraction, spectral flux, beat detection, and onset detection, and pretrained models for objective evaluation, e.g. VGGish, PaSST, CLAP. How compelling would these features be? Are there other features that would be useful to add?

Hi! If none is working on it, I would like to work on the models.

Hi @pablf, sorry for the late reply. That sounds great. Which models would you like to work on?

I have been working on the CLAP model. I am hoping to upload a draft within a few days, so we can discuss it.

@pablf sounds good — looking forward to it