How to calculate Zero Crossing Rate with PyAudio stream data

Question

I am trying to make to make some speech detection code. I am trying to do this with zero crossing rate. I did some research and found that when someone is speaking, the zero crossing rate should be a medium kinda value, not too high and not too low, but when I speak into the microphone the zero crossing rate becomes higher than it was with just background noise(which there is barely any) this is what I am doing to calculate it right now.

((audioData[:-1] * audioData[1:]) < 0).sum()

audioData is a numpy table and it's content is the result of pyAudioStream.read() could anyone tell me the correct way to calculate this? Thanks

Answer 1

That's a lot of unnecessary multiplication. Using a Boolean comparison and running it through np.diff will probably be faster:

zero_crosses = np.nonzero(np.diff(audioData > 0)))[0]

What this is doing:

1. creates a boolean array of where the signal is above 0 ( audioData > 0 )
2. does a pairwise difference ( np.diff ) so locations of zero crossings become 1 (rising) and -1 (falling)
3. picks the index of the array where those nonzero values are ( np.nonzero ).

Then if you want the number of crossings, you can just take zero_crosses.size .

As a bonus you have the timings of all the crosses so you can do things like a histogram that shows where more crosses are happening in your time history.