Audio-to-Audio
Stable Audio Tools
Safetensors
English

Will the causal version also be released?

#1
by stp99 - opened

In the paper, the authors state that the model is non-causal. However, they also mention training a causal version.

Will the causal variant be released as well?

The causal results in the paper were added in response to a request by the reviewers. We don’t currently have plans to release the causal model, as the design is not really optimised for the use-cases where causality is needed (like streaming).

If you want to explore a causal version yourself, it’s relatively easy to fine tune the released model with modified sliding windows and causal convolutions.

Thanks!

stp99 changed discussion status to closed
Your need to confirm your account before you can post a new comment.

Sign up or log in to comment