*CREPE+HYBRID TRAINING* A very experimental fork of the Retrieval-based-Voice-Conversion-WebUI repo that incorporates a variety of other f0 methods, along with a hybrid f0 nanmedian method.
MIT License
996
stars
215
forks
source link
Tensor size error and poor quality when performing a batch inference #226
For some reason, when I perform a batch conversion, the audio quality is God awful, and I get multitude of tensor error size errors.
Performing conversions on individual files yields great results, but no matter how much I play with my inferencing settings for batch conversions, I can't match the quality of an individual conversion. Does anyone know of a fix for this?
For some reason, when I perform a batch conversion, the audio quality is God awful, and I get multitude of tensor error size errors.
Performing conversions on individual files yields great results, but no matter how much I play with my inferencing settings for batch conversions, I can't match the quality of an individual conversion. Does anyone know of a fix for this?