search

csteinmetz1 / auraloss

Collection of audio-focused loss functions in PyTorch
Apache License 2.0
745 stars 67 forks source link

Expose mel-bin size parameter for multi-resolution STFT #61

Open csteinmetz1 opened 1 year ago

csteinmetz1 commented 1 year ago

From Zach: "pass in different mel-bin sizes for the multi-resolution STFT loss in auraloss. The new Improved RVQGAN paper shows that to be necessary."

image https://arxiv.org/abs/2306.06546