Summary: 1
To be published in the proceedings of IJCNN, Beijing, 1992
Context­Dependent Connectionist Probability Estimation in a
Hybrid HMM­Neural Net Speech Recognition System
Horacio Franco + , Michael Cohen + , Nelson Morgan # ,
David Rumelhart § and Victor Abrash +
+ SRI International, 333 Ravenswood Ave., Menlo Park, CA 94025
# Intl. Computer Science Inst., 1947 Center Street, Suite 600, Berkeley, CA 94704
§ Stanford University, Dept. of Psychology, Stanford CA 94305
Abstract
In this paper we present a training method and a network achitecture for the estimation of context­dependent observa­
tion probabilities in the framework of a hybrid Hidden Markov Model (HMM) / Multi Layer Perceptron (MLP)
speaker independent continuous speech recognition system. The context­dependent modeling approach we present
here computes the HMM context­dependent observation probabilities using a Bayesian factorization in terms of
scaled posterior phone probabilities which are computed with a set of MLPs, one for every relevant context. The pro­
posed network architecture shares the input­to­hidden layer among the set of context­dependent MLPs in order to
reduce the number of independent parameters. Multiple states for phone models with different context dependence
for each state are used to model the different context effects at the begining and end of phonetic segments. A new
training procedure that ``smooths'' networks with different degrees of context­dependence is proposed in order to
obtain a robust estimate of the context­dependent probabilities. We have used this new architecture to model general­

Source: Abrash, Victor - Speech Technology & Research Laboratory, SRI International

Collections: Computer Technologies and Information Sciences