摘要 |
PROBLEM TO BE SOLVED: To perform proper phoneme recognition to voiced sound and unvoiced sound respectively by varying spectral difference according to speech power of a frame. SOLUTION: Difference parameters set by a parameter setting process 11 for setting difference parametersα,βfor noise elimination are changed as necessary by a parameter changing process 12 according to speech power of a frame being object for a spectral difference outputted from a feature extraction part. The parameter changed by the parameter changing process 12 is inputted to a voiced/unvoiced sound judging process 13, and judged whether it is voiced sound or unvoiced sound as follows. The voiced/unvoiced sound judging process 13 judges the frame as a voiced sound part when the speech power of the frame changed in the parameter is larger than a predetermined threshold, and the frame is judged as an unvoiced part when the speech power is smaller than the threshold.
|