论文部分内容阅读
该文提出一种基于Gauss混合模型(GMM)托肯配比相似度校正得分(GMM token ratio similarity based score regulation,GTRSR)的说话人识别方法。基于GMM-UBM(通用背景模型)识别框架,在自适应训练和测试阶段计算并保存自适应训练语句和测试语句在UBM上使特征帧得分最高的Gauss分量编号(GMM token)出现的比例(配比),然后在测试阶段计算测试语句和自适应训练语句的GMM托肯分布的配比的相似度GTRS,当GTRS小于某阈值时对测试得分乘以一个惩罚因子,将结果作为测试语句的最终得分。在MASC数据库上进行的实验表明,该方法能够使系统识别性能有一定的提升。
This paper proposes a speaker recognition method based on GMM token ratio similarity based score regulation (GTRSR). Based on the GMM-UBM (universal background model) recognition framework, adaptive training statements and test sentences are calculated and saved in the adaptive training and testing phase. The ratio of appearance of GMM token with highest feature frame on UBM Ratio), and then calculate the matching similarity GTRS between the test sentence and the adaptive training sentence GMT tokens in the test phase. When the GTRS is less than a certain threshold, the test score is multiplied by a penalty factor, and the result is used as the final test sentence Score. Experiments on the MASC database show that this method can improve the system identification performance.