Score regulation based on GMM Token Ratio Similarity for speaker recognition

摘要

A novel approach named GTRSR (GMM Token Ratio Similarity based Score Regulation) for speaker recognition is presented in this paper, which judge the reliability of a test score based on GMM Token Ratio Similarity. GMM Token which is the index of the UBM component giving the highest score is saved for each frame during the training and test phase. Then the amount for each GMM Token is added up to form a vector GTR which stands for the GMM Token ratio of an utterance. In the test phase, we compute the similarity between the GMM Token ratio of test utterance and training utterance for a target speaker, i.e. GTRS. When GTRS is smaller than a threshold, the original likelihood score is regulated by multiplying a penalty factor as the final score of this test utterance. Experiments conducted on MASC@CCNT show our GTRSR can improve the performance of speaker recognition.