Improved HMM models for high performance speech recognition

In this paper we report on the various techniques that we implemented in order to improve the basic speech recognition performance of the BYBLOS system. Some of these methods are new, while others are not. We present methods that improved performance as well as those that did not. The methods include Linear Discriminant Analysis, Supervised Vector Quantization, Shared Mixture VQ. Deleted Estimation of Context Weights, MMI Estimation Using "N-Best" Alternatives, Cross-Word Triphone Models. While we have not yet combined all of the methods in one system, the overall word recognition error rate on the May 1988 test set using the Word-Pair grammar has decreased from 3.4% to 1.7%.