Toward Evolution Strategies Application in Automatic Polyphonic Music Transcription using Electronic Synthesis

TL;DR

This paper introduces a novel application of evolution strategies for automatic polyphonic music transcription, demonstrating improved accuracy and reduced computation time through parameter tuning and parallelization.

Contribution

It is the first to apply evolution strategies to music transcription, showing advantages over previous evolutionary algorithms in convergence speed and computational efficiency.

Findings

ES achieves fast convergence with proper parameter tuning

Parallelization reduces overall computation time

Transcription accuracy is comparable to existing methods

Abstract

We present in this paper a new approach for polyphonic music transcription using evolution strategies (ES). Automatic music transcription is a complex process that still remains an open challenge. Using an audio signal to be transcribed as target for our ES, information needed to generate a MIDI file can be extracted from this latter one. Many techniques presented in the literature at present exist and a few of them have applied evolutionary algorithms to address this problem in the context of considering it as a search space problem. However, ES have never been applied until now. The experiments showed that by using these machines learning tools, some shortcomings presented by other evolutionary algorithms based approaches for transcription can be solved. They include the computation cost and the time for convergence. As evolution strategies use self-adapting parameters, we show in this paper that by correctly tuning the value of its strategy parameter that controls the standard deviation, a fast convergence can be triggered toward the optima, which from the results performs the transcription of the music with good accuracy and in a short time. In the same context, the computation task is tackled using parallelization techniques thus reducing the computation time and the transcription time in the overall.