search

lmnt-com / diffwave

DiffWave is a fast, high-quality neural vocoder and waveform synthesizer.
Apache License 2.0
767 stars 112 forks source link

This can be used for voice conversion right? #53

Open dillfrescott opened 9 months ago

dillfrescott commented 9 months ago

As in, train the vocoder just on one speaker, and when you inference, even with another speaker, it will sound like the one you trained on?

dillfrescott commented 9 months ago

Ok update, I trained it on a specific person and when I tried inferencing with an unseen person it sounded too much like the actual unseen person. I would have thought it would sound like the speaker it was trained on.

Would switching it to unconditional mode help?