Landmark-Based Speech Recognition.pptVIP

Landmark-Based Speech Recognition Mark Hasegawa-Johnson Jim Baker Steven Greenberg Katrin Kirchhoff Jen Muller Kemal Sonmez Ken Chen Amit Juneja Karen Livescu Srividya Mohan Sarah Borys Tarun Pruthi Emily Coogan Tianyu Wang Goals of the Workshop “Acoustic Modeling:” Manner change landmarks: 15 binary SVMS Place of articulation: currently 33 binary SVMs, dependent on manner (t-50ms,…,t+50ms) “Lexical Modeling:” Dictionary implemented using current version of GMTK “Streams” in the dictionary: settings of lips, tongue blade, tongue body, velum, larynx Dependencies in GMTK learn the synchronization among the five articulators. Evaluation: Lattice rescoring (EARS RT03) Improve 1-Best WER Landmark-Based Speech Recognition Outline Scientific Goals of the Workshop Resources Speech data Acoustic features Distinctive feature probabilities: trained SVMs Pronunciation models Lattice scoring tools Lattices Preliminary Experiments Planned Experiments Scientific Goals of the Workshop Acoustic Learn precise and generalizable models of the acoustic boundary associated with each distinctive feature, … in an acoustic feature space including representative samples of spectral, phonetic, and auditory features, … with regularized learners that trade off training corpus error against estimated generalization error in a very-high-dimensional model space Phonological Represent a large number of pronunciation variants, in a controlled fashion, by factoring the pronunciation model into distinct articulatory gestures, … by integrating pseudo-probabilistic soft evidence into a Bayesian network Technological A lattice-rescoring pass that reduces WER Data Resources: Speech Data Data Resources: Acoustic Features Background: a Distinctive Feature definition Place of Articulation:cued bythe WHOLE PATTERN of spectral change over time within 150ms of a landmark Software Resources: SVMs trained for binary distinctive feature classification Software Resources: Posterior probability estimator based o