基于全局背景模型和竞争者模型的说话人确认系统

大多数说话人确认系统都设置一个背景模型用于描述假冒者的特性。文章提出一种新的说话人确认的背景模型,对所有说话人采用同一全局背景模型(简称UBM),并为每个说话人建立一个竞争者模型(cohort model)。在全局背景模型不能做出准确判断的情况下,启用竞争者模型再次进行判决。该模型充分利用了传统全局背景模型和竞争者模型的互补性。实验表明新的背景模型使系统性能有较明显的提高。

Speaker verification based on UBM and cohort model

Most comprehensive speaker verification systems use background speakers to model imposters.Based on analysis of general background speaker models,a new model combining the universal background model(UBM) with a cohort model is proposed.In the proposed speaker verification system,the same UBM is set for all claimed speaker,and a cohort model is set for each speaker.When the UBM fails to give a definite decision,the cohort model is used to make a verdict.The new background speaker model takes advantages of both UBM and the cohort model.Experimental results show that an equal error rate lower than conventional techniques is obtainable.