摘要 |
PURPOSE: To efficiently obtain a voice with excellent tone quality by small operation quantity by vector-quantizing a feature parameter when a difference of feature parameters between plural frames is large, and when the difference is small, vector-quantizing the difference of the parameters. CONSTITUTION: A subtractor 10 subtracts the locally decoded feature parameter vector of a preceding frame from the input vector Xi=(x1 , x2 ,..., XN) of an i-th frame including a feature parameter as an element and inputs a difference vector Yi to a vector quantizer 4, and the size of the vector Yi is compared with a threshold. When the vector Yi is smaller than the threshold, the vector Yi is compared with plural difference code vectors Z stored in a dictionary 3 and a code vector Zj having the shortest distance from the vector Yi is selected. When the vector Yi is larger, the input vector Xi is compared with the code vectors stored in the dictonary 3 and the code vector Zj having the shortest distance from the input vector Xi is selected. Since the vector-quantized signal is used, a voice with excellent tone quality can be obtained. |