Implementing F0 mean filtering

[tovaru]

The original repository added this new feature that greatly improves audio quality for songs with a lot of harmonics and reverb. I think it would be great to have it here, would make things much easier for getting accurate voices on songs.

[BlueAmulet]

Relevant commits
svc-develop-team/so-vits-svc@9eb0525
svc-develop-team/so-vits-svc@f25e566

Note: This doesn't seem to actually be of much besides a selector between parselmouth and crepe.

[Lordmau5]

On first glance, it seems like this project already does handle separate F0 methods / calculations

Namely, "pyworld" for dio and harvest, "crepe" for crepe and crepe-tiny, and "parselmouth" for parselmouth

https://github.com/34j/so-vits-svc-fork/blob/main/src/so_vits_svc_fork/f0.py#L188

I am, however, not sure if these can be improved with the changes in those commits

[34j]

As usual, I'm not sure what the upstream repo code is doing.

Why not simply median().bfill()?