An adaptive technique for automated recognition of musical tones

Abstract

This paper discusses the design of a software system that accepts as input a digitized waveform representing an acoustical music signal, and that attempts to derive the notes from the signal so that a musical score can be produced. The signal processing algorithms involved include event detection, or the discovery of precisely where within the signal the various notes actually begin and end, and pitch extraction, or the identification of the pitches being played in each interval. The paper presents the details of the algorithm, and compares the computer-generated score with the actual score for a song played with three instruments.