High-Accuracy Large-Vocabulary Speech Recognition Using Mixture Tying and Consistency Modeling

HIGH-ACCURACY LARGE-VOCABULARY SPEECH RECOGNITION USING MIXTURE TYING AND CONSISTENCY MODELING Vassilios Digalakis and Hy Murveit SRI International Speech Technology and Research Laboratory 333 Ravenswood Ave., Menlo Park, CA 94025-3493 ABSTRACT Improved acoustic modeling can significantly decrease the error rate in large-vocabulary speech recognition. Our approach to the problem is twofold. We first propose a scheme that optimizes the degree of mixture tying for a given amount of training data and computational resources. Experimental results on the Wall Street Journal (WSJ) Corpus show that this new form of output distri- bution achieves a 25% reduction in error rate over typical tied- mixture systems. We then show that an additional improvement can be achieved by modeling local time correlation with linear discriminant features. 1. INTRODUCTION To improve the acoustic-modeling component of SRI's DECI- PHER TM speech recognition system, our research has focused on two main directions. The first is to decrease the degree of mix- ture tying in the mixture observation densities, since conrinuous- density hidden Markov models (HMMs) have recently been shown to outperform discrete-density and tied-mixture HMMs [16]. The second is the removal of the simplifying output inde- pendence assumption commonly used in HMMs. Tied mixtures (TM) achieve robust estimation and efficient com- putation of the density likelihoods. However, the typical mixture size used in TM systems is small and does not provide a good representation of the acoustic space. Increasing the number of the mixture components (the codebook size) is not a feasible solution, since the mixture-weight distributions become too sparse. In large-vocabulary problems, where a large number of basic HMMs is used and each has only a few observations in the training data, sparse mixture-weight distributions cannot be esti- mated robustly and are expensive to store. To solve this problem, we follow the approach of simultaneously reducing the code- book size and increasing the number of different sets of mixture components (or codebooks). This procedure reduces the degree of tying, and the two changes can be balanced so that the total number of component densities in the system is effectively increased. The mapping from HMM states to codebooks can be determined using clustering techniques. Since our algorithm transforms a "less" continuous, or fled-mixture system, to a "more" continuous one, it has enabled us to investigate a number of traditional differences between tied-mixture and fully continu- ous HMMs, including codebook size and modeling of the speech features using multiple vs. single observation streams. Our second main research direction is focused on removing the simplifying assumption used in HMMs that speech features from different frames are statistically independent given the underly- ing state sequence. In this paper we will deal with the modeling of the local temporal dependencies, that is, ones that span the duration of a phonetic segment. We will show through the use of recognition experiments and information theoretic criteria that achieving decorrelation of the speech features is not a sufficient condition for the improvement in recognition performance. To achieve the latter, it is necessary to improve the …

Content truncated to 500 words