Emulation of Human Rhythm Perception

Abstract

Most music is said to a beat. When people describe music in this way, they mean that they can tap their feet to it, or dance to it, or somehow indicate their perception of some regularly spaced emphasis. In recent years there has been a growing interest in emulating this process with a computer program, so that computers, too, can understand how music is organized into beats. One reason such a program is desirable is that it would greatly enhance the ability of computers to participate intelligently in live musical performances. Another reason is that the program would enable computers to intelligently transcribe performed music, which would make music editing programs much easier to use. Furthermore, the problem is interesting from the point of view of psychologists of music, since perception of rhythm is one of the most basic activities of musical cognition (Bamberger 1980). Efforts so far to develop computer programs that are capable of parsing music rhythmically have been largely unsuccessful. The main reason for this, I believe, is that researchers (myself included) tend to underestimate how difficult this problem is. The rhythm of a musical performance is usually obvious to us, which causes us to think that writing a computer program to detect rhythm will be easy. We forget that the mental machinery that detects rhythm has been evolving for a long time, and has apparently become quite sophisticated. Emulating this sophistication on a machine is a demanding task. Despite these difficulties, the effort described below has produced some encouraging results. In what follows I will present a more precise exposition of the problem, and then describe Fa, a working rhythm-parsing program that runs on an Apple Macintosh computer. Basic Terms