A Genetic Search Technique for Polyphonic Pitch Detection.
Guillermo García · University of Michigan Library Repository · 2001
We present a method for estimating multiple, simultaneous fundamental frequencies in a polyphonic signal spectrum. The method takes advantage of the power of genetic algorithms to explore a very large search space with numerous local optima, and searches for a globally optimal combination of fundamental frequencies that best explains the polyphonic signal. No a-priori assumption about the number of fundamental frequencies is made, and no upper bound on this number needs to be imposed. Simulations show very good performance on signals with up to six fundamental frequencies, and performance remains acceptable when detecting up to eight fundamental frequencies.