Artificial musician. (Musical Instrument Digital Interface, Artificially Intelligent Computer Performer) (Column)
by Steven Anzovin
The advent of MIDI (Musical Instrument Digital Interface), an interface that lets electronic instruments talk to a computer, has revolutionized music composition. With a computer and a few MIDI-controlled instruments, one composer can write and perform what in the preelectronic era required an entire orchestra. But live performers who sit on the jam have to follow the computer, since the computer is unable to follow them. That stifles a human performer's style and makes for mechanical-sounding music.
In a new research project called AICP (Artificially Intelligent Computer Perforner), artificial intelligence (AI) researcher and amateur violist Bridget Baird is trying to reverse that situation--to create a computer system that can listen to and follow along with one or more live performers.
AICP exists only in the former whaling town of New London in a lab at tiny Connecticut College, where Baird is a professor. (Baird's address is Department of Mathematics, Connecticut College, Mohegan Avenue, New London, Connecticut 06320. Her E-mail address is bbba@conncoil.bitnet.)
Three years ago, Baird, fellow mathematician Donald Blevins, and music professor Noel Zahler came up with the idea of a program that could play a synthesizer to accompany MIDI input from an instrument played by a live musician. Similar projects have been undertaken at MIT and Carnegie-Mellon.
By last year, AICP could expertly follow one human player--as long as he or she played only a string of single notes. The program, running on a Macintosh equipped with a MIDI interface, "listens" to a person playing another MIDI instrument--for example, a keyboard or a MIDI-miked violin--and uses artificially intelligent rules to figure out how to stick to the score while keeping up with the live performer. If the live performer changes tempo or key in a way not specified in the score, AICP can still follow along without missing a beat. AICP does depend on a score, so improvision isn't within its capabilities. Yet.
Last summer, Baird snagged a National Science Foundation grant to enhance AICP to follow multiple players. "Usually the NSF tosses any proposal right into the trash can if it has the words art or music in it," says Baird, "but to my surprise, they gave us the money." Since no single processor could handle the input, Baird turned to parallel processing. She installed several Inmos Transputer boards in the Mac, one to handle MIDI from each live performer, thereby obtaining the processing power of a mainframe for a few thousand dollars (though the temperamental Inmos and Mac operating systems crash each other daily). Three students--Miriam Fendel, who is a bassoonist and psychology major; Chris Amorossi, who did the coding; and Dave McClendon, who designed the interface--got AICP to respond to as many as eight players.
AICP raises larger philosophical questions. "The main challenge," says Baird, "is to figure out what people are really doing when they make music together." For example, Fendel, the psychology major, is trying to figure out how musicians choose between conflicting inputs--whether to follow the first violin, who might be off-tempo, or the first flute, who might be off-key. AICP must learn to make the same decisions. Baird notes, "As programmers we have to verbalize what we are doing as musicians and then crystallize that into computer code."
Down the road, it should be possible to turn AICP into a music tutor, a professional-level accompanist, or maybe even a French horn in a first-rank orchestra. It should even be possible to make an AICP-controlled instrument sound like it's being played by a famous virtuoso--keyboard by Vladimir Horowitz or Jerry Lee Lewis--whose style could be reduced to an algorithm. So far, there's nothing like AICP in the world of consumer music software. Baird says she'd like to see AICP become a commercial product someday, but without more money (her NSF grant will run out at the end of the summer of 1992), that's unlikely to happen. For now, string trios looking for a fourth will just have to hire a real live person. Given the current dearth of jobs for human musicians, maybe that's not such a bad thing.