| 摘要 |
PROBLEM TO BE SOLVED: To reduce the amount of computations required for voice recognition by forcibly silencing a phoneme result. SOLUTION: Voice data from a voice inputting section are A/D converted in an A/D conversion process 11 and supplied to a recognition process starting discrimination process 12. If it is discriminated that recognition is being processed in the process 12, comparison discrimination takes place between the threshold value used in a silence/ sound discrimination process and the power of an inputted voice frame in a comparison discriminating process 13. Having obtained the result of the discrimination in the process 13 and if the power of the voice frame is larger than the threshold value of silence/sound, a feature extraction/phoneme recognition process is executed, the output of the process is inputted into a phoneme recognition computing process 14 and a phoneme recognition result is outputted. If the power of the voice frame is smaller than the threshold value of a silence/sound in the result of the discrimination of the process 13, the voice frame is judged to be silence in a silence discrimination process 16 and no feature extraction/phoneme recognition process is conducted, the recognition result is forcibly set to silence and inputted into a matching process executing process 15. |
