Joint Colloquium Distinguished Lecture Series

Getting an Earful: Speech Processing Research in Berkeley

photo of Nelson Morgan Wednesday, November 5, 2008
306 Soda Hall (HP Auditorium)
4:00 - 5:00 pm

Nelson Morgan
Professor in Residence
Electrical Engineering and Computer Sciences, UC Berkeley, and
Director, International Computer Science Institute


Over the last 50 years, speech processing has been a significant application for signal processing, statistical modeling, and artificial intelligence, and has become a distinctive research area of its own. During that time, speech technology (e.g., recognition, synthesis, or coding) has sometimes moved from a purely academic topic to everyday use by consumers. Despite this historical progress, there are still many problems for which current techniques fail badly. In many cases, these failures occur for tasks that are trivial for the natural speech processing systems of human beings.

In this talk, I will describe a range of problems tackled by the Speech research group at the International Computer Science Institute (ICSI), a neighboring institution closely affiliated with the Department. I will conclude with my own current obsession, which is the development of speech recognition methods that incorporate representations that are inspired by measurements in mammalian brains. The goal of these methods is to significantly improve recognition accuracy under noisy and reverberant acoustic conditions. Their principal drawback is an apparent explosion of the computational requirements, but this need has led to a collaboration with others in the Berkeley ParLab.

(Note: some of this material is covered in EE 225D, which I will be teaching in the Spring).


Nelson Morgan is the Director of the International Computer Science Institute (ICSI), an independent not-for profit research laboratory that is closely affiliated with UC Berkeley. In addition to directing the Institute he has led the Speech Group at ICSI since 1988. He is also is a Professor-in-residence in the EECS Department at the University of California at Berkeley, where he received his Ph.D. as an NSF Fellow in 1980. He has been working on problems in signal processing and pattern recognition since 1974, with a primary emphasis on speech processing. He is a former Editor-in-chief of Speech Communication, and has been a member of the IEEE Speech Technical Committee and the IEEE Neural Networks Committee. He is also a Fellow of the IEEE. In 1997 he received the Signal Processing Magazine best paper award. He was the Principal Investigator for the multi-site coalition funded by the DARPA EARS Novel Approaches project, which was the 2002-5 US government program focusing on long term progress in speech recognition.

  Return to EECS Joint Colloquium