Improving Speech Recognition Performance via Phone-Dependent VQ Codebooks and Adaptive Language Models in SPHINX-I1 M. Hwang R. Rosenfeld E. Thayer R. Mosur L. Chase R. Weide

This 'paper presents improvements in acoustic and language modeling for automatic speech recognition. Specifically, semi-continuous HMMs (SCHMMs) with phonedependent VQ codehooks are presented and incorporated into the SPHINX-IIspeech recognition system. The phonedependent VQ codebooks relax the density-tying constraint in SCHMMs in order to obtain more detailed models. A 6%' error rate reduction was achieved on the speakerindependent 20,000-word Wall Street Journal (WSJ) task. Dynamic adaptation of the language model in the context of long documents is also explored. A maximum entropy framework is used to exploit long distance trigrams and trigger effects. A 10% -- 15% word error rate reduction is reported on the same WSJ task using the adaptive language modeling technique.