Metaheuristic optimization of multiple fundamental frequency estimation

Abstract

Multiple fundamental frequency estimation is an important research area of the Music Information Retrieval (MIR). It can be used to transcribe polyphonic music from audio signal to symbolic data (e.g. MIDI files). The correct behaviour of some algorithms heavily depends on correct choice of a certain set of parameters. In this work, both the most fundamental and the most recent approaches in multiple F0 estimation task are presented. Furthermore, basic information about metaheuristic optimization is introduced. Finally, the developed solution is presented – it is the modified version of one of the most popular approaches – iterative algorithm – optimized using a popular example of metaheuristic optimization, i.e. Luus-Jaakola approach. The most important parameters are selected and their influence on the final results is discussed.