Development of Speaker Recognition Systems

Development of Speaker Recognition Systems
Thursday, August 6, 2015, 9:45 AM -10:45 AM
Presenter: Dr. Pedro A. Torres-Carrasquillo
Location: Newbury


Recent gains in the performance of automatic speaker recognition systems have been obtained by new methods in subspace modeling. This talk presents the development of speaker recognition systems ranging from traditional approaches, such as Gaussian mixture modeling (GMM) to novel state-of-the-art systems employing subspace techniques, such as factor analysis and iVector methods. This seminar also covers research on the means to exploit high-level information. For example, idiosyncratic word usage and speaker-dependent pronunciation are high-level features for recognizing speakers. These high-level features can be combined with conventional features for increased accuracy. The seminar presents new methods to increase robustness and improve calibration of speaker recognition systems by addressing common factors in the forensic domain that degrade recognition performance. We describe MIT Lincoln Laboratory’s VOCALINC system and its application to automated voice comparison of speech samples for law enforcement investigation and forensic applications. The talk concludes with appropriate uses of this technology, especially cautions regarding forensic-style applications, and a look at this technology’s future directions.

Presenter Bio

Dr. Pedro A. Torres-Carrasquillo joined the Information Systems Technology Group, currently Human Language Technology Group, at MIT Lincoln Laboratory in July 2002 as a technical staff member. At MIT Lincoln Laboratory, he has been involved in a number of areas related to information extraction from speech, including language, dialect and speaker recognition. His current areas of interest include automatic dialect recognition by combining multiple knowledge sources and speaker recognition in low data conditions focusing on forensic speaker recognition. Recently, he has been overseeing the development of software tools for speaker recognition.  He was co-chair for the IEEE Odyssey 2006 Workshop in San Juan, Puerto Rico. He is a Senior Member of the IEEE and is currently a member of the Speech and Language Technical Committee (SLTC).  From March 1995 to June 1998, he was an instructor at the Polytechnic University of Puerto Rico, where he is still an adjunct professor.  Dr. Torres-Carrasquillo received the B.S. degree from the University of Puerto Rico at Mayaguez in 1992, the M.S. degree from Ohio State University in 1995 and the Ph.D. degree from Michigan State University in 2002 all in electrical engineering.

Share thisShare on Google+Share on LinkedInTweet about this on TwitterShare on Facebook