| 摘要 |
PROBLEM TO BE SOLVED: To provide a speech processor which is improved in the speech recognition rate of speech recognition processing and increased in processing speed by compressing a speech with a small calculation quantity by providing pattern tables by men and women. SOLUTION: A VQ(vector quantization) pattern table 8 previously contains a VQ pattern table (made voice) 8a and a VQ pattern table (female voice) 8b which are sectioned by prescribed voice frequency energy. When a speech is inputted, a speech gender discrimination part 5 compares and analyzes the frequency energy sum of prescribed calculated frequency band groups of the input speech to discriminate the voice gender, and with a changeover switch 8c, the VQ pattern table corresponding to the speech gender discrimination result is selected. A minimum error pattern matching part 9 compares the similarity between a speech pattern normalized by a normalization part 4 and the selected VQ pattern table and outputs a speech compression result. |
