| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To realize speaker recognition of high precision when a speaker is recognized. <P>SOLUTION: A DP matching part 11 finds the optimum matching series F whose total of phonetic distances is minimum by usingΔpitch time series of two feature parameter series A and B and finds the total of individual distances, by using the optimum matching series and cepstrum coefficient time series of the two feature parameter series A and B, an inter-speaker distance calculation part 12 finds the total of the individual distances, and an identification part 13 identifies a speaker based on the total. Consequently, phone-decomposing performance and speaker-decomposing performance are compatible and stable recognizing performance can be secured, so high-precision speaker recognition can be realized. <P>COPYRIGHT: (C)2006,JPO&NCIPI</p> |
