This thesis addresses the problem of multiple fundamental frequency estimation in a chord. The problem is complicated because real instrument sounds are never pure sinus oscillations, but combinations of those. Exploration of sound and also our problem solution are based on the mathematical concept of Fourier analysis.
We developed an idea and implemented our own algorithm for multiple fundamental frequency estimation. Algorithm and its working principle are made for basic chord recognition. Algorithm is based on a construction of a simple model for a single instrument. In its basic form it works in a controlled environment. We tested our solution on our own sound recordings of a digital piano, digital organ, and acoustic (concert) accordion. In the cases of digital piano and organ, we reached 75 and 94 percent success of recognition (on 100 different chords). On the accordion, our algorithm fully recognized 49 of 50 different chords.
From the obtained results we conclude that our idea and algorithm are correct and applicable. Results and performance of our algorithm even surpass capabilities of some skilled musicians.
|