Recent Progress in the Sphinx Speech Recognition System

RECENT PROGRESS IN THE SPHINX SPEECH RECOGNITION SYSTEM Kai-Fu Lee, Hsiao-Wuen Hon, Mei-Yuh Hwang Computer Science Department Carnegie Mellon University Pittsburgh, PA 15213 Abstract This paper describes recent improvements in the SPHINX Speech Recognition System. These enhancements include function-phrase modeling, between-word coarticulation modeling, and corrective training. On the DARPA resource management task, SPHINX attained a speaker-independent word accuracy of 96% with a grammar (perplexity 60), and 82% without grammar (perplexity 997). 1. Introduction SPHINX is a large-vocabulary, speaker-independent, continuous speech recognition system based on discrete hidden Markov models (HMMs) with LPC-derived parameters. In order to deal with the problem of speaker independence, we added knowledge to these HMMs in several ways. We represented additional knowledge through the use of multiple codebooks. We also enhanced the recognizer with word duration modeling. In order to model co-articulation in con- tinuous speech, we introduced the use of function- word-dependent phone models, and generalized triphone models. More recently, we have made considerable progress with the SPHINX System. We reformulated the general- ized triphone clustering algorithm as a maximum- likelihood procedure, and carried out some experiments with generalized triphones. We also implemented and evaluated the modeling of function phrases, and between-word coarticulation modeling rising general- ized triphones. The latter experiment reduced SPHINX's error rate by 24-44%. We modified the corrective train- ing algorithm [1] for speaker-independent, continuous speech recognition. Corrective training reduced SPHINX's error rate by 20-24%. In this paper, we will describe all components of the SPHINX System, with emphasis on the recent improve- ments. The SPHaNX System has been described in [2] and [3]. Publications on the recent improvements will be forthcoming. On the 991-word DARPA resource management task, SPHINX achieved speaker-independent word recog- nition accuracies of 82% and 96%, with grammars of perplexity 991 and 60, respectively. Results with the 1988 and 1989 test data resulted in 78 and 76% without grammar, and 96% and 94% with the word pair gram- mar. 2. Speech Representation The speech is sampled at 16 KHz, and pre- emphasized with a filter of 1 - 0.97z -1. Then, a Ham- ming window with a width of 20 msec is applied every 10 msec. Autocorrelation analysis with order 14 is fol- lowed by LPC analysis with order 14. Finally, 12 LPC- derived cepstral coefficients are computed from the LPC coefficients, and these LPC cepstral coefficients are transformed to a mel-scale using a bilinear trans- form. These 12 coefficients are vector quantized into a codebook of 256 prototype vectors. In order to incor- porate additional speech parameters, we created two ad- ditional codebooks. One codebook is vector quantized from differential coefficients. The differential coef- ficient of frame n is the difference between the coef- ficient of frame n+2 and frame n-2. This 40 msec. difference captures the slope of the spectral envelope. The other codebook is vector quantized from energy and differential energy values. 3. Context-Independent HMM Training SPHINX is based on phonetic hidden Markov models. We identified a set of 48 phones, and a hidden Markov model …

Content truncated to 500 words