| 摘要 |
PURPOSE:To improve the recognition accuracy by selecting a pulse sound source or noise sound source according to distribution information on high and low frequency components, correcting variation between adjacent data so that the variation decreases, and then modulating the signal of the sound source with a standard or featured pattern and composing an audible sound. CONSTITUTION:The pulse sound source 12 and noise sound source 13 are provided and used selectively according to the distribution information on the high and low frequency components of a voice, and then variation between respective adjacent data in one group of data on the frequency base of the standard pattern or featured pattern is corrected by using low-pass filters F1 - F15 so that the variation becomes small. Then the signal of the sound source is modulated with the pattern and the audible sound is composed 11. Consequently, a voice can be reproduced by adding information of the sound source which is lost once approximately, so how to voice a voice in its registration and the registration contents are remembered and the recognition accuracy is improved. |
