报 告 人：Eric Fosler-Lussier 美国俄亥俄州立大学教授，长期从事语音识别、机器学习方面的研究
主 持 人：谢磊教授
报告题目：Beyond the linear chain: explorations of structured discriminative models for automatic speech recognition
Hidden Markov Models have formed the backbone of speech recognition techniques for decades: their power lies in the simplicity of the basic model, which is adaptable to many different types of training regimes and representations of hidden states. The current state of the art utilizes both discriminative observation models, in the form of Deep Neural Networks, as well as discriminative training criteria. In this talk, I will describe a line of research from my lab that has incorporated different kinds of linguistic and structural representations in a direct discriminative model, following a Conditional Random Field approach to modeling. I will highlight several points of development of these models: initial phone recognition models that integrate phonological and phonetic information, factored phonological transcription models, factored discriminative training methods, and segmental models that incorporate information over long time spans. We find that introducing structures that can represent interesting temporal and linguistic events we can construct novel competitive alternative models for speech recognition. (Joint work with Jeremy Morris, Rohit Prabhavalkar, Preethi Jyothi, and Ryan He.)
Eric Fosler-Lussier is an Associate Professor of Computer Science an Engineering, with a courtesy appointment in Linguistics at The Ohio State University. After receiving a B.A.S. (Computer and CognitiveScience) and B.A. (Linguistics) from the University of Pennsylvania in 1993, he received his Ph.D. in 1999 from the University of California, Berkeley, performing his dissertation research at the International Computer Science Institute under the tutelage of Prof. Nelson Morgan. He has also been a Member of Technical Staff at Bell Labs (2000-2002), Lucent Technologies, a Visiting Researcher at Columbia University (2003), and a visiting scholar at the Nisonger Center, A University Center for Excellence in Developmental Disabilities at Ohio State (2010-11).
Widely published in both speech and language processing, research recognitions include the National Science Foundations CAREER award (2006), the Ohio State College of Engineering Lumley Research Award (2010), and the 2010 IEEE Signal Processing Society Best Paper Award (2010, co-authored with Jeremy Morris). He served as program chair for the North American Association for Computational Linguistics Meeting (2012, with Ellen Riloff and Srinivas Bangalore), and has served two terms on the IEEE Speech and Language Technical Committee. He is a member of the Association for Computational Linguistics, the International Speech Communication Association, and a senior member of the IEEE.