Mon, Oct 08, 2012 @ 01:30 PM - 03:00 PM
Ming Hsieh Department of Electrical Engineering
Conferences, Lectures, & Seminars
Speaker: Andrew Rosenberg, Queens College (CUNY)
Talk Title: Noisy Text -- Techniques and Tools for Prosodic Analysis
Abstract: Prosody is a crucial component to human spoken communication. A number of phonological models of prosody have been proposed to facilitate empirical testing of linguistic hypotheses. The incorporation of categorical/ symbolic prosodic information into spoken language processing systems remains limited. A more common approach to incorporating prosodic information into spoken language processing applications has been the direct incorporation of acoustic/prosodic features into a feature vector.
This talk will discuss issues around symbolic modeling of prosody including some reasons for why direct modeling has been more prolific than symbolic modeling. Particular focus will be given to recent improvements to AuToBI, a toolkit for automatic ToBI labeling, in terms of feature representation, classification approaches, and software engineering. I'll also cover some recent applications of prosodic analysis to spoken language processing tasks including pronunciation modeling, keyword search, and classification of speaking style and nativeness.
Biography: Andrew Rosenberg is an Assistant Professor of Computer Science at Queens College (CUNY). His research concerns spoken language processing and machine learning. He received his Ph.D. from Columbia University in 2009. His dissertation, and much of his research has focused on techniques and applications of automatic prosodic analysis on which he has written over 30 papers. Andrew Rosenberg also contributed to the IBM Jeopardy! Challenge, working with the speech synthesis team to improve Watson’s voice.
Host: Prof. Shrikanth Narayanan
Audiences: Everyone Is Invited
Posted By: Mary Francis