April 28, 2004

The microprocessor alone could have the potential to change music forever – how we make it and how we hear it. But instead of innovating new musical processes, composers, scientists and engineers caught themselves in a rut of emulation, almost frightened by the possibilities of eventually limitless memory, bandwidth and algorithmic complexity. Instead of working with the physical and expressive constraints of the machines as we would with any other instrument we ended up building systems that detect “World” music or developing software that can recreate a plucked string in a steel cage.

“Computer music” should not be “fast composition–” a human composer’s belief propagated through their model 1.2 billion times a second, it should be the residuals of a machine’s listening and expressive capabilities. It should be something a human could never create, not even slowly. Over the past five years I’ve tried to teach computers about music on their own so they can grow musical intelligence and classify and compose for themselves. I’ll present recent work on this “music acquisition,” and software and hardware implementations for music retrieval and synthesis convolving around the seven tenets of music to computers.

http://web.media.mit.edu/~ bwhitman
http://eigenradio.media.mit.edu/

Bio:
Brian Whitman is a computer scientist and musician who works with the intersection of sound processing and machine learning to create musically intelligent systems for performance, retrieval and synthesis. He has been involved in the field of music retrieval since 1999, first creating the “Minnowmatch” startup at NEC Research Institute, which offered new signal processing and statistical learning approaches to music recommendation, and later consulting for audio firms and software synthesis companies. He received his Master of Science degree from Columbia University studying Natural Language Processing and is currently a PhD candidate at the MIT Media Lab in the Music, Mind and Machine group, where his research concerns “music acquisition–” how humans and computers learn about music from observation through unsupervised learning and language and signal processing. His synthesis and interactive music projects investigate the peculiar constraints and temperament that machine learning systems exhibit when asked to be creative.