My purpose is to silence all the parts of a .wav audio where there is no speech. I am currently using webrtcvad, but what I achieve is just removing the non-speech part from the audio (with their example.py code: https://github.com/wiseman/py-webrtcvad/blob/master/example.py). If someone can point me or show me a how to achieve my goal, I would be grateful! This sounds also sounds like a background noise removal problem.
1 Answers
Assuming that you want the WAV output to have the same duration as the input, just with the non-speech areas being replaced with silence, and the speech areas unaltered.
The way to do this is to multiply the audio signal with the output from the detector. The detector should output 1.0 for passing though (speech signal), and 0.0 for silencing (non-speech).
Sometimes one uses a small value instead of 0.0 for the blocking part, to just reduce the volume a bit, without making it pure silence. For example 0.01 (-20 dB).
Sometimes an abrupt transition can be a bit rough. In this case one may apply a bit of smoothing or fade. A simple alternative is an exponential moving average.
Related