Segment-Based Acoustic Models for Continuous Speech Recognition

Mari Ostendorf, Jan Robin Rohlicek

Abstract: The goal of this project is to develop improved statistical models for speaker-independent recognition of continuous speech, together with efficient search algorithms appropriate for use with these models. The current work on acoustic modeling is focussed on: stochastic, segment-based models that capture the time correlation of a sequence of observations (feature vectors) that correspond to a phoneme; hierarchical stochastic models that capture higher level intra-utterance correlation; and multi-pass search algorithms for implementing these more complex models. In addition, we have extended the effort on models of high order statistical dependence to language modeling. This research has been jointly sponsored by ARPA and NSF under NSF grant IRI-8902124 and by ARPA and ONR under ONR grant N00014-92-J-1778.