Building Noise Robust Machine Listeners with Data and Inspiration from Humans

Michael Mandel

Assoc. Professor

Brooklyn College, Graduate Center

November 8, 12pm-1pm

Room N919

Matching human performance is one of the most difficult problems for a variety of speech communication technologies, including automatic speech recognition, voice processing in hearing aids, and mobile telephony.  One theory of human noise robustness is that listeners pick out reliable “glimpses” of a target sound and utilize contextual clues to fill in missing information using top-down knowledge.  This talk presents work that brings both of these processes to machines.