How does it work?
The Function Call
1
1.) the input signal is taken and segmented into different blocks in which each block has a window function (hamming window) applied to it so that minimal spectral leakage occurs
2.) Each block is then zero-padded and the fft of each block is computed separately
3.) The frequency bins of each of the segmented block is calculated using the indices and the sampling frequency of 44,100Hz
4.) Depending on the given Frequency, the blocks representing that narrowband region of frequencies is extracted and a surface plot is made for those audio signals as seen below.
2
The plots on the left is compared with the plot below
3
To validate the audio signal we take the cross correlation of the two signals. (In this case we used Ground truth1 and recording1)
It can be seen that at that point (x=35) is when the two signals are correlated the most, meaning if we shift of the signals that amount of times, they will resemble each other the most. However this must be shifted back 33
