Using cross correlation to find similarity in two audios. Useful in echo cancellation.

find_similarities.py

import audio_io
import numpy

audio1, _ = audio_io.ReadWavFile('voice_nodelay.wav')
audio1 = audio1.ravel()

audio2, _ = audio_io.ReadWavFile('voice_mic_blend3.wav')
audio2 = audio2.ravel()

# Truncate all signals same length, then pad to avoid boundary effects.
n = min(
    audio1.shape[0],
    audio2.shape[0])

audio1_padded = numpy.zeros((2 * n, ))
audio2_padded = numpy.zeros((2 * n, ))

audio1_padded[:n] = audio1[:n]
audio2_padded[:n] = audio2[:n]

# Use cross-correlation to estimate the impulse response of the room
# and speakers.
print( 'FFT(A)')
A = numpy.fft.fft(audio1_padded)
print( 'FFT(B)')
B = numpy.fft.fft(audio2_padded)
print( 'conj(A)')
Bconj = numpy.conj(B)
print( 'IFFT(A*B^)')
xcorr = numpy.fft.ifft( numpy.multiply(A, Bconj) )

audio_io.WriteWavFile(xcorr[:2*n], 44100, 'xcorr2.wav')