Title: Markovian Combination of Language and Prosodic Models for better Speech Understanding and Recognition

Abstract: Traditionally, "language" models capture only the word sequences of a language. A crucial component of spoken language, however, are its rhythmic and melodic properties, i.e., its prosody. This talk will summarize recent work on integrated, computationally efficient modeling of word sequences and prosodic properties of speech, for a variety of speech recognition and understanding tasks, such as dialog act tagging, disfluency detection, and segmentation into sentences and topics. In each case it turns out that hidden Markov representations of the underlying structures and associated observations arise naturally, and allow existing speech recognizers to be combined with separately trained prosodic classifiers. The same HMM-based models can be used in two modes: to recover hidden structure (such as sentence boundaries), or to evaluate speech recognition hypotheses, thereby integrating prosody into the recognition process. More information about individual research projects, as well as the publications listed, are available at http://www.speech.sri.com/projects/hiddenevents.html, http://www.speech.sri.com/projects/sieve/, and http://www.clsp.jhu.edu/ws97/discourse/.

Curriculum: Andreas Stolcke received his undergraduate degree in Computer Science from the Technische Universitaet Munich in 1988, and a Ph.D. in Computer Science from the University of California at Berkeley in 1994. He was a research assistant and postdoctoral researcher at the International Computer Science Institute (ICSI) in Berkeley, doing research on connectionist and probabilistic methods for natural language processing. His doctoral thesis investigated learning and parsing algorithms for probabilistic grammars. Andreas is currently a Senior Research Engineer with the Speech Technology and Research Laboratory at SRI International, as well as a Visiting Researcher at ICSI. His recent work has been on statistical models for speech recognition and understanding, focusing on language and prosodic models for spontaneous and conversational speech.

Stolcke.ps Stolcke.pdf