Featured Alum: Dan Ellis

Dan Ellis is an associate professor at Columbia University in New York City. His research focus is signal processing and machine learning techniques, which he applies to speech recognition and music analysis. He received his PhD in electrical engineering at MIT. Upon completion of his degree, he worked at ICSI as a postdoctoral researcher from 1996-1999, then as a staff research scientist before being offered his current position at Columbia in 2000. Ellis continues to maintain an active affiliation with ICSI. Since leaving ICSI, he has worked with ICSI's Speech Group on the EARS project as well as on speech processing for meetings, and he is involved in this issue's featured project, GALE.

At ICSI, Ellis researched signal processing techniques, with a particular focus on using signal processing for automatic speech recognition (ASR). Through his ASR work he had the opportunity to collaborate with many different people, including Professors Nelson Morgan and Hervé Bourlard. He says that collaboration was one of the most valuable things he learned at ICSI. Its success at enhancing the quality of research there. This instilled in him a strong belief in the value of collaborative research, and his continued cooperation with ICSI and other institutions has had a strong influence on his career thus far. He feels that understanding how to make collaborative projects work is a valuable skill as an academic.

Although speech technology continues to be a big part of Ellis's research, he recently began some work using some of the same signal processing and machine learning techniques to identify features in music. In one current project, he trains an algorithm to recognize chords in music. Chords are a basic unit for music, analogous to the words in speech. Although there is much more work to do, such as increasing the amount of training data and defining which features are important, the results are promising so far. He hopes to develop applications for music signal processing, such as information retrieval, creating customized playlists for users, and determining new music that a user might like based on its similarity to the user's known preferences. Ellis teaches courses on music signal processing and machine learning at Columbia, in addition to general signal processing and speech and audio signal processing.

Combining his affinity for collaboration as well as teaching, Ellis has been working with two of ICSI's graduate student researchers, Arlo Faria and Kofi Boakye. Faria contacted Ellis while updating the Quicknet software originally developed at ICSI by Ellis and Dave Johnson. It is being used as front end for ASR in the GALE project, and Faria is adapting it to work more quickly on ICSI's current server set-up. Ellis is informally advising Boakye as he works on his PhD in Electrical Engineering at UC Berkeley. Ellis enjoys working with Boakye, as they have very similar research interests, and hopes to collaborate with him on future projects.

Late-Breaking Alumni News

Since conducting our interview with Dan Ellis, we learned that his family has grown to include a new baby girl, Zoe, born on February 6th. Congratulations to Dan and his family!

Jeff Bilmes and Katrin Kirchoff, both Speech Group alums, also have a new baby. Their son Alexander was born on March 2nd. Jeff and Katrin, who met while working at ICSI, are now on the faculty of University of Washington, where they continue to pursue speech-related research. Congratulations to Katrin and Jeff!