Deep Neural Networks for Speech Processing

Alex Acero

Microsoft Research

Wednesday, February 27
2:00 PM, Lecture Hall

Abstract

Neural Networks are having a renaissance thanks to a mathematical formulation known as Restricted Boltzmann Machines and the availability of more processing power. Unlike past learning algorithms for neural networks, it is now possible to learn the weights for networks with many layers, and thus the names "deep neural networks" and "deep learning": In this talk I'll describe this new formulation and its application to speech recognition. Deep neural networks have resulted in significant error rate reductions in speech recognition and other fields. I’ll explain what I think is different from past incarnations of neural networks.

Bio

Alex Acero is the Director of the Conversational Systems Research Center at Microsoft Research, Redmond. He manages a team conducting research on audio, speech, and language processing. His team has contributed to Microsoft products such as Kinect. He has also managed research groups in machine translation, information retrieval, multimedia signal processing and computer vision. Dr. Acero is also an affiliate Professor of Electrical Engineering at the University of Washington. He received a M.S. degree from the Polytechnic University of Madrid in 1985, a M.S. degree from Rice University in 1987, and a Ph.D. degree from Carnegie Mellon University in 1990. Prior to joining Microsoft in 1994, Dr. Acero worked at Apple and Telefonica.

Dr. Acero is a Fellow of IEEE and ISCA. He is President-Elect for the IEEE Signal Processing Society. Dr. Acero is author of the books Spoken Language Processing (Prentice Hall, 2001) and Acoustical and Environmental Robustness in Automatic Speech Recognition (Kluwer, 1993), has written invited chapters in 6 edited books and over 225 journal and conference papers. He holds 110 US patents.