Improvement of speaker recognition system by individual information weighting

In speaker recognition, it is very important to use individual information extracted from speech waves. Most of the speaker recognition methods assume that each part of speech has equal amount of information to represent a speaker, although it differently contribute to speaker recognition. The aim of this paper is to suggest a new scoring method of the HMM, which applies different importance to all the basic portions of a sampled speech waveform. we first define the quantity of the importance of speech frames, propose how to measure it and apply to speaker recognition. The performance of the proposed method was compared to non-weighting HMM based speaker recognition system. In speaker verification experiments, the proposed method reduced equal error rates considerably as compared to a conventional method which treats all speech segments to have the same importance. In speaker identification experiments, the proposed method marked relatively 28% higher recognition rate than the baseline system, and was more robust in long-term variation. These results demonstrate that the proposed method is efficient in measuring speaker information and more appropriate for speaker recognition.