Current Research

I am currently employed as a speech scientist at SpeechCycle, a company that specializes in automated spoken dialog systems for trouble-shooting applications. My day-to-day activities comprise tuning rule-based grammars, building statistical semantic models, and overseeing transcribers and annotators. Other long-term research investigations that I am heading include the development of web-based tools for continuous performance analysis and visualization, a real-time automatic detector of unsuccessful calls, and automated caller satisfaction monitoring.

Resent Academic Research

I received my Ph.D. from Columbia University on June, 2007. My dissertation is entitled, “Prosody and Speaker State: Paralinguistics, Pragmatics, and Proficiency” and centers on the computational modeling of acoustic-prosodic information for use in the automatic detection of several types of speaker state, which I define to be any long- or short-term cognitive process. In particular, I explored the usefulness of prosody in the prediction of emotion in three domains—a speech-enabled Intelligent Tutoring System, a call-center application, and a corpus of acted emotion—and found that prosody is utilized differently depending not only on the emotion in question but also the domain. Notable was the finding that modeling the temporal change of prosodic information, such as pitch and loudness, improves detection accuracy. I also found prosody to be useful for detecting student questions in Intelligent Tutoring Systems and for automatic proficiency scoring of non-native English speakers.

For other research I have conducted in the past, please see my Curriculum Vitae and my publications.