search

lucidrains / voicebox-pytorch

Implementation of Voicebox, new SOTA Text-to-speech network from MetaAI, in Pytorch
MIT License
589 stars 49 forks source link

Apply mask to cond #11

Closed stevenhillis closed 1 year ago

stevenhillis commented 1 year ago

I think the mask wasn't actually being used to mask frames of cond. The mask was being used to restrict loss computation only to "masked" frames, but the frames of cond weren't actually being masked.

lucidrains commented 1 year ago

@stevenhillis yes indeed, thanks Steven! i gather deepgram is now interested in voicebox?