摘要 |
PROBLEM TO BE SOLVED: To obtain a high recognition performance even through memory capacity and throughput of a CPU are largely restricted by vector-quantizing an input speaker code book by using a universal code book, and generating a quantized input speaker code book. SOLUTION: By using each mapping function and a 1st-a 3rd specified speaker group code books CB1, CB2, CB3 (each code book size is 256), respective input code books of the speaker #1, the speaker #2 and the speaker #3 are generated. A universal code book 94 is a code book obtained from a wide class including speakers from adults to children, and has a large size such as 1024 or 2048. Thus code books (256-size) using only the code numbers of the universal code book 94 are generated by vector-quantizing input code books by this universal code book 94. In such a manner, a data quantity can largely be decreased.
|