| 摘要 |
PURPOSE: To perform high quality coding with a low bit rate without increasing the number of outgoing bits. CONSTITUTION: A noise coding book 51 -5N is made from a learned voice for each range in which a pitch period of a voice is classified, an input voice is analyzed in linear prediction, a quantization predictive coefficient is set to a coefficient of a synthetic filter 3, segmented with a pitch period from an immediately preceding filter excitation vector in an adaptive coding book 4, and one on the noise coding book 51 -5N is selected in accordance with the pitch period. Weighted sum of the noise vector and the segmented adaptive coding vector is made to be an excitation signal of the filter 3 from the noise coding book 51 -5N, and the pitch period and the noise vector are selected so that distortion of a synthesized speech is minimized. |
