| 摘要 |
PURPOSE:To obtain a high recognition rate for a general speaker on a voice recognition method that performs the recognition employing standard patterns made of connecting voice pieces such as CV(consonant+vowel).VC(vowel+ consonant) and VCV(vowel+consonant+vowel). CONSTITUTION:A voice pieces dictionary.DP pass form storage section 7 stores the degree of similarity obtained by matching the voices generated by few speakers and the standard patterns for general speakers stored in a phoneme standard patterns storage section 4, the patterns obtained by segmenting voice pieces patterns from the time sequence of the regression coefficient vectors and the information of DP pass forms set for every frame employing average continuation length of the phonemes in each voice piece, distribution, minimum and maximum lengthes. In a recognition section 10, the parameter series which are similarly made from input voices and the recognition objective dictionary made by connecting the voice pieces dictionary are DP-matched by changing DP pass forms for every frame and the recogintion result is obtained by making the dictionary item which has the largest degree of similarity. |
