The researchers demonstrated their methods by encoding a 20-second clarinet solo in less than a single kilobyte.

The technique involved isn't an audio recording technology; rather, it re-creates the clarinet solo in the same way that a player piano re-creates a piano piece from a roll of punched paper. But in addition to re-creating the notes, it also re-creates the way in which the player played the notes.

Mark Bocko, professor of electrical and computer engineering and co-creator of the technology, suggests that perhaps the future of music performance lies in reproducing performers rather than recording them.

Bocko and doctoral students Xiaoxiao Dong and Mark Sterling programmed a computer to model clarinet fingering, breath pressure, and lip pressure and to use that information to affect the sound described by their model of a virtual clarinet. Postdoctoral researcher Gordana Velikic and Dave Headlam, a professor of music theory at the University of Rochester, also contributed to the research.

By using its programmed knowledge of clarinets and clarinet players, Bocko's approach avoids having to sample music thousands of times a second, which generates a lot of data and makes music files large.

Bocko expects that his team's work will lead to more expressive electronic and computer-generated music. He also anticipates that the technology could be extended to generate vocals and voices more naturally.

The technique isn't perfect yet, but Bocko expects his music synthesis algorithms will become more accurate.