RVC-Project
Noise in silent parts after voice conversion
▼Issue
Generation of noise in silent parts after voice conversion.
▼Work flow
- I created a learning dataset of 300 files by removing silence and separating them using Audacity based on high-quality voice sound sources.
- I converted it from high-quality reading sound source.
- Intense noise occurs in the silent part.
The training settings were batch size 24, and I tried epoch 50, 200, 500, etc., but it seems to occur consistently.
▼Sample
voice:JPN (children book recitation )
▼Environment
- RVC-beta.7z(20230428updated)
- Windows10 22H2
Same here, but one strange thing, I trained two models with two datasets. One model has this silent noise issue while the other one does not.
One model has this silent noise issue while the other one does not.
Me too. I have a model that uses a different data set and is less noisy.
The dataset I used to train the model that I posted as a sample had very high quality voices, but there was a little background music, so I removed the background music with Demucs-GUI and split the audio. However, the cause seems to be something else, as I have also generated models that are noisy even on fully voice-only datasets.
Also, I tried post-processing with noise gate, etc., but I couldn't cut it because the dB of the voice and noise were about the same.
This issue (Noise in silent part) seems to have been resolved with the latest update. great job! thank you.