search

openvpi / DiffSinger

An advanced singing voice synthesis system with high fidelity, expressiveness, controllability and flexibility based on DiffSinger: Singing Voice Synthesis via Shallow Diffusion Mechanism
Apache License 2.0
2.62k stars 275 forks source link

Is removing background noise from audio beneficial to the quality of DiffSinger? #190

Closed Alistair-zhong closed 1 month ago

Alistair-zhong commented 1 month ago

Thank you very much to the OpenVPI team for maintaining the open-source DiffSinger branch.

I want to train a DiffSinger Model, and wanna know Is removing background noise from audio beneficial to the quality of DiffSinger?

image

Thanks~

yqzhishen commented 1 month ago

The noise shown in your screenshot does not seem severe. You can just crop the silence segments.

blueyred commented 1 month ago

@Alistair-zhong depends on how prominent the noise is, say you have some classic microphone hum at a constant 180 or 300Hz - your singing model will learn it and it will be present in your generations. If its something thats very occasional you may find it gets buried by your other recording data. Generally though the audio you put in you get out.