VAD on short audio packet

[eziolotta]

Nice project !!
From my test, rVADfast performs good segmentation of a medium sized wav file ( > 10/15 seconds )
However, If we cut the input audio (ex: to 5 sec. ), the results are not the same, why is a posteriori analysis done?

It would be nice to apply this library in realtime systems, where server analyzes audio stream, that is short packets of bit-audio in sequence.

[eziolotta]

This is my experiment with rVADFast:
MozillaItalia/DeepSpeech-Italian-Model#107 (comment)

[HolgerBovbjerg]

Hi,

For now, we are not considering streaming VAD use.
Not sure what is meant by, "why is a posteriori analysis done?". However, you can find explanations for the various design choices in rVAD in the corresponding research paper: https://arxiv.org/abs/1906.03588