Add VAD feature for faster-whisper

[dev-jesser]

In faster-whisper there's a built-in VAD feature. It's set to default to false here

This feature should be helpful to handle instances which I believe are called "hallucinations" - the speaker goes silent for some time, then picks up again, but the resulting text is just repeating the last sentence before the person went silent.

Can a parameter be added to the REST API to optionally set this vad_filter feature to true and add vad_parameters?

[blundercode]

I also agree this feature would be useful, VAD seems to be the standard now for whisper and would love to see it used here.