| 摘要 |
PURPOSE:To make high-quality voice encoding possible by causing correlation to be exist between the time characteristic of a quantization error and the time characteristic of inputted voices when a standard pattern is selected. CONSTITUTION:Easily encodable feature quantities are extracted from voice patterns divided into frames of regular intervals by means of an analyzing section 20 and time changing characteristics are extracted from analyzed features by means of a section 30 for extracting the time changing characteristic of inputted voices. Then a code vector searching section 50 selects the optimum code vector by using the code book inputted from a code book 40 holding previously fixed standard patterns as code vectors and the time changing characteristic of the reproduced sound pattern calculated from the time changing characteristic inputted from the section 30 and the index of previously selected code vectors accumulated in a memory 60. Therefore, voice encoding which is free from deterioration becomes possible. |
