All I really need to know I learned from Lou and Nat: Nat Durlach

Abstract

Nat Durlach was one of my two primary mentors in graduate school at MIT. In the context of my PhD thesis, Nat taught me the invaluable lesson of how to think. In this talk I will describe how I applied to Speech Recognition what I learned from him about the value of reviewing prior literature, how to write a strong research proposal, and the need to question basic assumptions. These are eternal gifts Nat bequeathed to me, and I will highlight them in a series of applications using examples over my career at IBM. Specifically, I will draw from work in making advances in core speech recognition, the creation of the interdisciplinary multi-site NSF MALACH project on providing access to large spoken archives of speech, and building one of the early Speech Recognition systems for Mandarin. In addition, I will also describe applications of our work in Clear speech to a set of speech recognition related problems, including the issue of “Sheep and Goats”—why speech recognition works well on some speakers but not ...