Abstract:
Microphones and headsets are everywhere.  In order to use them to provide real-world next-generation services, such as conversational mobile software agents, more detailed and context-dependent models of speech are necessary. This talk presents two projects towards this goal, motivated by the fact that human abilities to understand speech in noise far outstrip current automatic approaches.  It will first describe a new data-driven approach to building a context-dependent model of human speech perception in noise.  By formulating intelligibility prediction as a classification problem, the model can learn the important spectro-temporal features of speech utterances from listening test results. This is in contrast to traditional, heuristic approaches, which identify phonetic landmarks by hand.  The talk will then describe the use of context-dependent models of speech to perform noise robust automatic speech recognition.  Posed as an optimization problem, this system finds the speech features that minimize a combination of the distance to reliable regions of the noisy observation and the negative log likelihood under a complete large vocabulary continuous speech recognizer. It reduces both speech recognition errors and the distance between the estimated speech and the original clean speech.

Bio:
Michael I Mandel earned his BSc in Computer Science from the Massachusetts Institute of Technology in 2004 and his MS and PhD with distinction in Electrical Engineering from Columbia University in 2006 and 2010 as a Fu Foundation School of Engineering and Applied Sciences Presidential Scholar. From 2009 to 2010 he was an FQRNT Postdoctoral Research Fellow in the Machine Learning laboratory at the Université de Montréal.  From 2010 to 2012 he was an Algorithm Developer at Audience Inc, a company that has shipped over 200 million noise suppression chips for cell phones.  He is currently a Research Scientist in Computer Science and Engineering at the Ohio State University where he recently received an Outstanding Undergraduate Research Mentor award.