Distinguished Guest Lecturer

A Model-Based Approach to Speech and Language Processing

John Makhoul
Chief Scientist, BBN Technologies
Chief Scientist, BBN Technologies

May 29 2009 9:30AM
260 Dreese Labs
All interested parties are welcome.
Refreshments will be served prior to the talk.

Abstract:

The last two or three decades have witnessed significant advances in speech and language technologies, thanks to a mathematically rigorous, model-based approach. In this approach, statistical models, specified by experts, are trained automatically from annotated data where only domain knowledge is needed to perform the annotations and can usually be supplied by non-expert labor. The resulting systems have achieved state-of-the-art performance, are highly robust in the face of degraded input, and have proven to be language-independent, requiring only annotated training data from a new language. Best results are obtained when the model designs are properly informed by structural and linguistic knowledge of the problem at hand.

This talk will present model-based approaches to three technologies: speech recognition, optical character recognition, and machine translation. In each case, we show how the model-approach has been used to advance the state of the art. The presentation will end with a demonstration of a real-time broadcast monitoring system where live broadcasts in Arabic and Mandarin Chinese are transcribed automatically and translated into English using off-the-shelf systems.

Bio:

Dr. Makhoul is one of the world's leading experts on speech and signal processing. He has been with BBN Technologies since 1970, working on various aspects of speech and language processing, including speech coding, speech synthesis, speech recognition, speaker identification and verification, artificial neural networks, digital signal processing, optical character recognition, language understanding, speech-to-speech translation, and human-machine interaction using voice. In addition, Dr. Makhoul is Adjunct Professor at Northeastern University where he supervises students doing their Ph.D. work at BBN in the area of speech and language processing.

Dr. Makhoul is a Fellow of the IEEE and a Fellow of the Acoustical Society of America. His 1975 IEEE Proceedings paper on linear prediction was named a "Citation Classic" by the Institute for Scientific Information. His honors include the 1978 Senior Award, the 1982 Technical Achievement Award, the 1988 Society Award of the IEEE Signal Processing Society, the IEEE Third Millennium Medal, and most recently, the recipient of the IEEE 2009 James L. Flanagan Speech & Audio Processing award "for pioneering contributions to speech modeling."

He is also an alumnus of The Ohio State University, having graduated with an MS from Ohio State in the Electrical Engineering department, and subsequently he received his Ph.D. from the Massachusetts Institute of Technology.

Host: John Fosler-lussier