Add VAD support

[satra]

files could have a lot of silence before or after. Add VAD functionality and add it optionally to the b2ai feature pipeline.

[GasserElbanna]

Some options to consider:

• https://huggingface.co/pyannote/brouhaha
• https://huggingface.co/pyannote/segmentation
• https://huggingface.co/speechbrain/vad-crdnn-libriparty

[fabiocat93]

A simpler implementation for removing silence from start and end of audio files is here:

• https://pytorch.org/audio/main/generated/torchaudio.transforms.Vad.html

Also, a note on pyannote-audio: since the last version, it's recommended to use the speaker-diarization pipeline and not the segmentation model because it works better in terms of diarization error rate