RI: Medium: Collaborative Research: Explicit Articulatory Models of Spoken Language, with Application to Automatic Speech Recognition

点击翻译按钮获取中文摘要

Proposal Title: RI: Medium: Collaborative Research: Explicit Articulatory Models ofSpoken Language, with Application to Automatic SpeechRecognitionInstitution: Toyota Technological Institute at ChicagoAbstract Date: 05/22/09This award is funded under the American Recovery and Reinvestment Act of 2009(Public Law 111-5).One of the main challenges in automatic speech recognition is variability in speakingstyle, including speaking rate changes and coarticulation. Models of the articulators(such as the lips and tongue) can succinctly represent much of this variability. Mostprevious work on articulatory models has focused on the relationship between acousticsand articulation, but more significant improvements require models of the hiddenarticulatory state structure. This work has both a technological goal of improvingrecognition and a scientific goal of better understanding articulatory phenomena.The project considers larger model classes than previously studied. In particular, theproject develops graphical models, including dynamic Bayesian networks andconditional random fields, designed to take advantage of articulatory knowledge. A newframework for hybrid directed and undirected graphical models is being developed, inrecognition of the benefits of both directed and undirected models, and of bothgenerative and discriminative training. The project activities include major extension ofearlier articulatory models with context modeling, asynchrony structures, andspecialized training; development of factored conditional random field models ofarticulatory variables; and discriminative training to alleviate word confusability.The scientific goal addresses questions about the ways in which articulatory trajectoriesvary in different contexts. Existing databases are used, and initial work in manualarticulatory annotation is being extended. In addition, the project uses articulatorymodels to perform forced transcription of larger data sets, providing an additionalresource for the research community. Other broad impacts include new models andtechniques with applicability to other time-series modeling problems. Extending theapplicability of speech recognition will help it fulfill its promise of enabling more efficientstorage of and access to spoken information, and equalizing the technological playingfield for those with hearing or motor disabilities.NATIONAL SCIENCE FOUNDATIONProposal AbstractProposal:0905633 PI Name:Livescu, KarenPrinted from eJacket: 06/10/09 Page 1 of 1