Segment-Based Acoustic Models for Continuous Speech Recognition.
Abstract
This research aims to develop new and more accurate stochastic models for speaker-independent continuous speech recognition by extending previous work in segment-based modeling, by introducing a new hierarchical approach to representing intra-utterance statistical dependencies, and by developing language models that capture topic dependencies. These techniques, which have high computational costs because of the large search space associated with higher order models, are made feasible through a multi-pass search strategy that involves rescoring a constrained space given by an HMM decoding. We expect these different modeling techniques to result in improved recognition performance over that achieved by current systems, which handle only frame-based observations and assume that these observations are independent given an underlying state sequence. The primary research efforts and results over the last two quarters have included implementation of several software system improvements to enable research in more general distribution clustering and score combination weight estimation; development of a constrained EM algorithm for training the mixture language model which led to a small improvement in performance over Viterbi-style training; development of a mixture version of cache language modeling, together with a new content- word cache model, obtaining a small error reduction for short (3-sentence articles); implementation and evaluation of three lattice search algorithms, providing an understanding of conditions under which the different algorithms are most appropriate. (AN)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 31, 1994
- Accession Number
- ADA289969
Entities
People
- J. R. Rohlicek
- Mari Ostendorf
Organizations
- Boston University