Basser Seminar Series

Audio Pro with Multiple DSPs and Dynamic Load Distribution

Barry Vercoe, MIT Media Lab

Wednesday 2 February 2005 2-3pm

Basser Conference Room (G92) Madsen Building


The latest professional Karaoke system released in Japan has no ASIC for sound synthesis and effects processing, but instead a small group of load-sharing DSPs that co-operatively handle the varied and dynamically varying tasks of complex high-quality audio performance. The software-only system is a first for the audio industry, heralding a new generation of downloadable and task-sensitive software that delivers time-critical performance from distributed general-purpose silicon. The tasks of emulating a 64-voice orchestra plus real-time MPEG decode, live voice tracking with pitch and tempo following, and a full range of audio effects processing are represented in a network of active objects which are just-in-time serviced by a co-operating array of SIMD DSPs.

This talk will begin with a general overview of research in the MIT Media Lab, along with some specific examples of sensing human expressive attributes while tracking live instrumental and vocal performance. A deeper account of the core technology of MPEG-4 audio will lead to the above case of how to amass the raw compute power to handle major professional audio challenges. We close with some pleasing examples of client-end musical results, and finally some insights into where the industry appears to be heading.

About the speaker

Barry Vercoe is Professor of Music and Professor of Media Arts and Sciences at MIT, and Assoc Academic Head of the Program in Media Arts & Sciences . He was born and educated in New Zealand in music and in mathematics, then completed a doctorate in Music Composition at the University of Michigan. In 1968 at Princeton University he did pioneering work in the field of Digital Audio Processing, then taught briefly at Yale before joining the MIT faculty in 1971. In 1973 he established the MIT computer facility for Experimental Music - an event now commemorated on a plaque in the Kendall Square subway station.

During the ‘70’s and early 80’s he pioneered the composition of works combining computers and live instruments. Then on a Guggenheim Fellowship in Paris in 1983 he developed a Synthetic Performer - a computer that could listen to other performers and play its own part in musical sync, even learning from rehearsals. In 1992 he won the Computer World / Smithsonian Award in Media Arts and Entertainment, and recently gained the 2004 SEAMUS Lifetime Achievement Award.

Professor Vercoe was a founding member of the MIT Media Laboratory in 1984, where he has pursued research in Music Cognition and Machine Understanding. His several Music Synthesis languages are used around the world, and a variant of his Csound and NetSound languages has recently been adopted as the core of MPEG-4 audio - an international standard that enables efficient transmission of audio over the Internet. At the Media Lab he currently directs research in Machine Listening and Digital Audio Synthesis (Music, Mind and Machine group), and is Associate Academic Head of its graduate program in Media Arts and Sciences.