variogr.am home | latest | writings | gallery

writings at variogr.am

On this section

Research statement

From 2001 until 2005 I taught computers how to hear music at the MIT Media Lab, working in the Music, Mind and Machine (formerly "Machine Listening") group with Barry Vercoe. Previously I was at the NEC Research Institute in Princeton working on Minnowmatch, a machine listening engine. Even earlier, I was working in the Natural Language Processing group at Columbia University.

I want to help people find music, and I want artists and labels to find people. Our work concentrates on both the content and culture of music through signal processing, data mining, language analysis and machine learning techniques. We can claim higher accuracy in music retrieval tasks such as recommendation and similarity over the currently-popular marketing influenced collaborative filtering approaches by looking at both the signal and the community response to the signal. I also work on learning a a query-by-description front end for music retrieval by correlating music with free text (community metadata.)

We can predict a community's reaction to a new piece of music well enough to create personalized automatic record reviews. This can level the field for independent under-marketed musicians by having a bias-free musical intelligence perform our filtering. Instead of "Rock and Pop" we can create meaningful similarity clusters of music that are tuned to individual tastes and styles.

My other research interests include generative perceptual synthesis, text retrieval and understanding, audio and visual scene analysis, time-aware kernel approaches to machine learning, and embedded sound hardware.

Short texts, interviews and summaries

Academic papers and talks