lucidrains / naturalspeech2-pytorch

Implementation of Natural Speech 2, Zero-shot Speech and Singing Synthesizer, in Pytorch
MIT License
1.28k stars 99 forks source link

Fix EMA ignore set #9

Closed amiasato closed 1 year ago

amiasato commented 1 year ago

Very minor bugfix. EMA intermediate results were almost pure noise and it was probably due to the ignore set being wrong - the set(str) constructor created a set with each char of the string instead of creating a set with a single string.

lucidrains commented 1 year ago

@amiasato oh shoot! thank you