Abstract

Spoken keyword search is a detection task to locate all occurrences of a lexical entry in an audio stream. Such a task is especially challenging for low-resource spoken languages, since mainstream solutions rely heavily on large quantities of manual transcriptions to train high performance automatic speech recognizers. In this talk, I will give an overview of the low-resource spoken keyword search framework developed at the Institute for Infocomm Research in the context of the NIST Open Keyword Search Evaluations. In particular, I will present selected highlights of the keyword search system, including submodular optimization data selection to maximize acoustic diversity through Gaussian component indexed N-grams to transcribe speech, exemplar acoustic models that exploit cross-lingual bottleneck features and kernel density estimation, and a phonology-augmented statistical transliteration framework to model out-of-vocabulary words of foreign origin. 

Biography

Nancy F. Chen received her Ph.D. from the Massachusetts Institute of Technology (MIT) in 2011. She is currently a scientist at the Institute of Infocomm Research (I2R), Singapore. Her research interests include spoken language processing for low-resource languages, keyword search, speech recognition, and computer-assisted language learning. Prior to joining I2R, she worked at MIT Lincoln Laboratory on her Ph.D. research, which integrates speech technology and speech science, with applications in speaker, accent, and dialect characterization. Dr. Chen has also helped organized conferences such as Odyssey 2012: The Speaker and Language Recognition Workshop and INTERSPEECH 2014. Dr. Chen is a recipient of multiple awards, including the Microsoft-sponsored IEEE Spoken Language Processing Grant, the MOE Outstanding Mentor Award, and the NIH Ruth L. Kirschstein National Research Service Award. For more information, please see: http://alum.mit.edu/www/nancychen