Replace webrtcvad by silero-vad

[Edresson]

Advantages of the new script against the old one:

1. Better quality in trimming nonspeech parts.
2. Support to all sample rates (I'm using the 8khz Silero-VAD model and just mapping the time stamps between all possible sample rates >= 8Khz).
3. Now is possible to trim just the beginning and end of the audio (parameter --trim_just_beginning_and_end)

[erogol]

Is there a test for it?

Also I guess you need to rebase for the checks

[Edresson]

Is there a test for it?

Also I guess you need to rebase for the checks

The test is already implemented in tests/aux_tests/test_remove_silence_vad_script.py. Given that the old script and the new one have basically the same params we don't need to change it :). I will rebase it.