This week I worked on improving pitch detection. I am going with Power Spectral Density function and peak detection within that. I have cleaned the signal up a little more, since I noticed many issues with spurious peaks and also the recording picking up harmonics.
I have passed the signal through a human vocal range band pass filter.
I have also thresholded based on amplitudes.
I had a meeting with Professor Stern’s team to talk about potentially improving it further. Their team has an approach that uses autocorrelation functions and then a histogram of peak differences to pick the best peaks. He suggested modifying which peaks I select.
So I still need to smooth out the signal, this job has proven more difficult than imagined. I will attempt to do more with better peak selection from the PSD plot.
All I have left to do is to improve the signal smoothing like such and to collect more statistics on the heuristics. We will make our final presentation very soon and also the poster for the poster session. Finally, I will place my scripts in my public AFS space so that our website can interface with it and query it.
I seem to be in a tight spot, but am hopeful things will be on time.