This week I continued to research implementation options and tools for pitch detection and rhythm detection. I have narrowed down the process for pitch detection to involve the FFT of the input audio in combination with Autocorrelation and either HPS (Harmonic Product Spectrum) or HPSS (Harmonic Percussive Source Separation). Both Autocorrelation and HPS are common approaches to pitch detection, but have their own issues with potential harmonic disruption and a limited minimum range of around 40 Hz respectively. By using the two together, I plan to implement a pitch detection algorithm that uses the harmonic approach of HPS (or HPSS) to allow the algorithm to be used across a wide of variety of instruments (each with their own harmonics) while using autocorrelation to overcome the frequency limits of HPS. For rhythm detection, I have found that there is ongoing research into this topic at CMU under Chris Donahue and have reached out to him regarding the methods I have looked into. The scope of approach is much more limited than Donahue’s, as we are restricting measurement to a specific tempo metronome in western classical marches, practice book exercises, and allegro’s instead of a general test against all sheet music, but would still like to learn more about Donahue’s work.
Even with the health issues I experienced this past week, I am still on schedule to continue research into this next week and finish our design proposal. This week I am going to be wrapping up my rhythm detection research, extend that into our music tracking algorithm research and implementation, and finalize these three main algorithms for our design proposal. I am also hoping to hear back from Chris Donahue.