Spaces:
Running
on
Zero
Running
on
Zero
Repeating reference audio
#1
by
flozi00
- opened
When using custom audio the model generates a Clone of the reference audio and ignores the Text given.
With sample audio and Text it Sounds good
The model is a bit undertrained (400k samples and no lr cooldown) and mainly handles clean audio like from the Emilia Pipeline. I’m open to adding the UVR-MDX-NET-Inst_HQ_3 ONNX model to the Space for easier cleanup. Since Emilia samples are working, I'd suggest trying a cleanup step or rising the temperature a bit.
Alright, increasing temperature is fixing it. great work, thanks for sharing
flozi00
changed discussion status to
closed