| 摘要 |
PURPOSE:To prevent the degradation of tone quality without extending a dictionary by using a binary BTSP (binary time spectrum pattern) to introduce a new parameter (voice power). CONSTITUTION:A voice synthesizing filter bank 6 corresponding to the center frequency of a BTSP channel is provided, and only a filter bank corresponding to the channel of a higher level of BTSP is driven. A ratio of the number of higher levels of BTSP on the low band side to that on the high band side is obtained, and the filter bank 6 is driven by a triangular wave string which has a power proportional to the voice power when this ratio is higher than a prescribed value. When this ratio is equal to or lower than the prescribed value, the filter bank 6 is driven by the white noise having the power proportional to the voice power. Thus, a recognition dictionary is made of binary data, and the tone quality is kept in a certain level even at the time of using it as a synthesizing dictionary. |
