| 摘要 |
PURPOSE:To prevent sound quality from being deteriorated by adding a white noise to residual information and using the added information for sound synthesis. CONSTITUTION:An analyzing part is provided with a means 14 for calculating a forecasting residual power Pe by using spectrum information Kn obtained from an input sound and a means for finding out a difference between the power Pe and a residual signal power Pq to be quantized to find out difference residual power Pw. On the other hand, a synthesizing part is provided with a means 36 for finding out the difference residual power Pw, spectrum information Kn and residual information gi, mi on the basis of a signal 35 supplied from the analyzing part, a means 40 for generating a white noise 42 from the power Pw, an adding means 45 for adding the noise 42 to the information gi, mi, and a synthesizing filter 47 for using the added result of the means 45 as a driving source signal 46 to synthesize sound from the information Kn. In a low bit rate in which the residual information can not be sufficiently transmitted, the error residual power is modeled by the white noise and natural synthesized sound having no roughness can be obtained. |
