|
|
MIT Electrical Engineering and Computer Science
EECS Event |
Monday, November 27, 2000
4:00 PM (refreshments 3:45)
Edgerton Hall, Room 34-101
EECS Colloquium
Abstract
This talk first highlights points in the history of hearing research at Bell Labs relevant to speech technology. It then discusses some of the new and emerging challenges and some recent activities designed to meet them.
From the 1920's to the 1940's, Fletcher and his colleagues conducted research in psychophysics of hearing (loudness, masking, etc.), and in speech perception, in the context of intelligibility (e.g. developing an Articulation Index). This work was driven by the need for knowledge to guide the development of the Bell System telephone network. From the 1950's through the 1970's, hearing research by Jim Flanagan (a doctoral alumnus who did a thesis in the speech group in RLE) and others at Bell Labs was focused on the psychophysics of hearing, supporting the development of speech coding systems (like the Vocoder).
In recent years major changes in network infrastructure are occurring, first in the Internet domain and now also in the Wireless domain. New hardware provides networks with much higher data rates. This allows transmission of wide-band speech and music, recorded with one or more microphones, with the emphasis on high quality reconstructed signals (i.e., transparent coding). Developing these new coding systems needs knowledge of how a human listener perceives such complex acoustic images.
Major advances have also occurred in man-machine dialogue systems. Applications now exist where, for limited tasks, machines achieve a reasonable performance in recognizing naturally spoken language while "conversing" with humans in a natural-sounding voice. Research in speech perception is needed to guide efforts to improve the naturalness of synthetic voice, and to achieve robust recognition given a variety of speakers and a noisy environment.