Abstract
Dynamic Programming, the Viterbi Algorithm, the theory of finite state machines, and the mathematics of Markov processes are well developed formalisms that can be applied to the problem of pattern matching acoustic events in speech. At the intersection of these formalisms are a set of computationally efficient but powerful algorithms that are a significant improvement over the commonly used pattern matching schemes for handling temporal and spacial deviation from idealized prototypes. This paper will describe some of these relatively new pattern matching algorithms. It will be shown that they reduce to simple forms that promote low cost isolated word and continuous speech recognition.

This publication has 6 references indexed in Scilit: