METHODS FOR RECOGNIZING CONTINUOUS SPEECH BY USING CONTINUOUS DYNAMIC PROGRAMMING, AND THEIR APPLICATION.

In continuous speech recognition using Continuous Dynamic Programming (CDP), it is necessary to perform post processing in order to determine the word output from the CDP values. The post processing is not of one type; the range extends from a simple type to a complex one which involves the introduction of syntactic constraints. Each method of post processing is thought to have a clear range of application from the standpoint of the vocabulary size and the discrimination rate. In experiments using the post processing method which makes a word lattice, there has not been any evidence of sharp drop in the discrimination rate without syntactic constraints for up to 200 words.