A new test data generation method for sequential probability ratio test

초록

In this paper, we propose a method that can generate new test data using the sample shift of starting frame for the sequential probability ratio test (SPRT) in automatic speaker verification (ASV). The SPRT is an effective algorithm that can reduce the computational complexity of test, but it's not suitable method to apply for the short utterance. The proposed method is effective in applying the SPRT regardless of the utterance length of the test data because it is a method to generate the new test data through the sample shift of starting frame. Further, it shows a dramatic result in decision of the threshold process. Experimental results show that the proposed method has 0.5% higher performance than a conventional method in equal error rate (EER), but introduce a little increase to the computational complexity. copyright © 2008 IEICE.