SYSTEM FOR LEARNING STANDARD PATTERN

摘要

PURPOSE:To characterize the standard patterns of respective categories and to contrive speaker adaptation with high accuracy by correcting standard patterns by using data for learning. CONSTITUTION:A 1st evaluation function to indicate the matching characteristic between a 1st parameter set and a 2nd parameter set indicating the data for learning is provided and a 2nd evaluation function consisting of the sum of the functions relating to the distances between the correction rates of the 1st parameter set weighted by the value indicating the matching characteristic between the 1st parameters is provided. The 1st parameter set is so determined as to optimize the value of the evaluation function created from the two evaluation functions. The pattern recognition having the high recognition performance without requiring a large amt. of pronunciation data is executed in this way.