search

Plachtaa / seed-vc

State-of-the-Art zero-shot voice conversion & singing voice conversion with in context learning
GNU General Public License v3.0
660 stars 76 forks source link

Issues with Streaming Live #53

Open asusdisciple opened 4 days ago

asusdisciple commented 4 days ago

First of all, great start for a beginning in VC! A very promising approach, I hope you will reiterate on. Inference already works pretty good besides the slightly worse quality than rvc v2. Streaming also works good but there are a lot of artifacts when you do not talk. So it can not process silence very good, as soon as you stop talking you hear chinese/english voice parts. I think in RVC they have a "silent" model called mute for this in the model directory, maybe this approach could help improve on the streaming capabilities.

Plachtaa commented 3 days ago

thanks for your kind advice, you are correct about streaming artifacts while not talking. I will add the fix to TODO list