NOISE ADAPTATION SYSTEM FOR VOICE MODEL, NOISE ADAPTATION METHOD, AND VOICE RECOGNITION NOISE ADAPTATION PROGRAM

摘要

PROBLEM TO BE SOLVED: To facilitate handling of a noise voice when an SNR varies and to suppress calculation cost, by generating a voice model in a single-tree structure and using it for voice recognition.;SOLUTION: All noise data in a noise database are used under all SNR conditions to calculate the distances between all noise models under the respective SNR conditions, and noise-superposed voices are clustered. According to the result of the clustering, a single-tree structure model space wherein noises and SNRs are integrated is generated (steps S1 to S5). In a feature extraction stage (step S6), an input noise voice to be recognized is analyzed, to extract a feature parameter series, and an optimum model is selected from the tree structure noise voice model space by likelihood comparison of HMMs (step S7). Linear conversion is so carried out that the likelihood is further maximized from the selected noise voice model space.;COPYRIGHT: (C)2005,JPO&NCIPI