摘要 |
The vector quantizer code book designing method using a majority representation vector splitting of a sound recognition system comprises the steps of obtaining a average distortion by a present representation vector from an initial representation vector and then obtaining a representation vector having a maximum average distortion; dividing a sound storage vector into a large collective representation vector and a small collective representation vector on the basis of the representation vector having the maximum average distortion and obtaining each centroid to increase the code book; obtaining an optimal representation vector and completing the formation of the code book when the number of the present representation vectors reaches a desired number; if the number does not reach the desired number, returning to the first step.
|