An approach of binary isomorphic quantization for speaker identification

Binary isomorphic quantization is a technique for reducing amount of feature vectors by determining their similar form. The feature vectors are extracted from speech. This method is based on a function that measures internal changing of feature vectors to produce binary vectors. The binary vectors are partitioned and then clustered the same binary vectors together. A set of clusters with the maximum frequency will be chosen to generate a codebook instead of using all binary vectors. An experimental results show the efficiency in speaker identification which gives high accuracy especially in the continuous speech. Moreover, we also investigate its performance by comparing it with other methods.