High-Performance Speech Recognition Using Consistency Modeling

Publisher Website
Google Scholar
Abstract
The goal of this project conducted by SRI International (SRI) is to develop consistency modeling technology. Consistency modeling aims to reduce the number of improper independence assumptions used in traditional speech- recognition algorithms so that the resulting speech-recognition hypotheses are more self-consistent and, therefore, more accurate. Consistency is achieved by conditioning HMM output distributions on state and observations histories, P(x/ s,H). The technical objective of the project is to find the proper form of the probability distribution P, the proper history vector, H, and the proper feature vector, x, and to develop the infrastructure (e.g. efficient estimation and search techniques) so that consistency modeling can be effectively used. During the first year of this effort, SRI focused on developing the appropriate base technologies for consistency modeling. We developed genonic hidden Markov model (HMM) technology, our choice for P above, and Progressive Search technology for HMM systems which allows us to develop and use complex HMM formulations in an efficient manner. Papers describing these two techniques are included in the Appendix of this report, and are briefly summarized below. This report also describes other accomplishments of Year 1, including the initial exploitation of discrete and continuous consistency modeling and the development of a scheme for efficiently computing Gaussian probabilities.

This publication has 0 references indexed in Scilit: