A Real-Time Pitch Recognition Algorithm for Music Applications
W.B. Kuhn · Computer Music Journal · 1990
A number of methods for transcribing audio input into a printed musical score have been described in the literature (Moorer 1977, Piszczalski et al. 1977). Most of this work has concentrated on the relatively difficult problem of transcribing music from polyphonic audio sources. A classic example is provided by Moorer (1977), who considered the difficulties involved in transcribing two guitars played simultaneously. In contrast, the simpler problem of transcribing a monophonic source (single voice) has received little attention. This neglect is unfortunate, since one of the most popular musical instruments, the human voice, is fundamentally monophonic. This paper describes a commercially viable technique for recognizing the pitch of the human voice in music applications. The method presented is capable of recognizing notes over a three-octave range or greater, in real-time, with a typical accuracy/ resolution of 1 percent (less than 20 cents). It has been implemented successfully in both hardware and software configurations. Applications for this technology include melody transcription, computer-assisted vocal instruction, games, and voiceto-MIDI conversion products.