[Bug] VITS model needs its own AudioConfig class

[erogol]

🐛 Description

Some of the fields in BaseAudioConfig are not used in VITS models and it is misleading.

This was also the reason for our silence issues when training our latest VITS models @Edresson

To Reproduce

You can set for instance silence trimming True but it would not work since VITS uses a custom load_audio function without
silence trimming.

Expected behavior

I've 3 solutions

1. VITSAudioConfig
2. Raise an error when a field in BaseAudioConfig is not being used (probably harder )
3. Find a smart way to override AudioProcessor instead of writing custom standalone functions.

[e0xextazy]

Has this problem been solved?