Improved Acoustic Modeling for Continuous Speech Recognition

IMPROVED ACOUSTIC MODELING FOR CONTINUOUS SPEECH RECOGNITION C.-H. Lee, E. GiachinP , L. R. Rabiner, R. Pieraccini and A. E. Rosenberg Speech Research Department AT&T Bell Laboratories Murray Hill, NJ 07974 ABSTRACT We report on some recent improvements to an HMM- based, continuous speech recognition system which is being developed at AT&T Bell Laboratories. These advances, which include the incorporation of inter-word, context-dependent units and an improved feature analysis, lead to a recognition system which achieves better than 95% word accuracy for speaker independent recognition of the 1000-word, DARPA resource management task using the standard word-pair grammar (with a perplexity of about 60). It will he shown that the incorporation of inter-word units into training results in better acoustic models of word juncture coarticulation and gives a 20% reduction in error rate. The effect of an improved set of spectral and log energy features is to further reduce word error rate by about 30%. We also found that the spectral vectors, corresponding to the same speech unit, behave differently statistically, depending on whether they are at word boundaries or within a word. The results suggest that intra-word and inter-word units should be modeled independently, even when they appear in the same context. Using a set of sub-word units which included variants for intra-word and inter-word, context-dependent phones, an additional decrease of about 10% in word error rate resulted. 1. INTRODUCTION In the past few years there have been proposed a number of systems for large vocabulary, speaker-independent, continuous speech recognition which have achieved high word recognition accuracy [1-5]. The approach to large vocabulary speech recognition we adopt in this paper is a pattern recognition based approach. The basic speech units in the system use phonetic labels and are modeled acoustically based on a lexical description of words in the vocabulary. No assumption is made, a priori, about the mapping between acoustic measurements and sub- word linguistic units such as phonemes; such a mapping t Now with CSELT, Torino, Italy. is entirely learned via a finite training set of utterances. The resulting speech units, which we call phone-like units (PLU's) are essentially acoustic descriptions of linguistically-based units as represented in ttle words occurring in the given training set. In the baseline system reported in [1], acoustic modeling techniques for intra-word context-dependent PLU's were discussed. The focus of this paper is to extend the basic acoustic modeling techniques developed in [1] to include modeling of word juncture coarticulation and to incorporate higher-order time derivatives of cepstral and log energy parameters into the feature vector in order to improve speech recognition performance. We tested the improved acoustic modeling techniques on speaker-independent recognition of the DARPA Naval Resource Management task using both the word-pair (WP) and the no grammar (NG) conditions. For the FEB89 test set using the WP grammar, the word accuracy improved from 91.3% to 95.0% when both the inter-word context-dependent PLU's and an improved feature analysis were incorporated into the baseline system. We also observed that, for the first time, over …

Content truncated to 500 words