话者识别中结合模型和能量的语音激活检测算法

Combination of Model and Energy Based VAD Algorithm in Speaker Recognition System

语音激活检测是检测语音起始终止端点的一种算法,合适地选择语音来进行说话人模型的注册和测试对话者识别系统的性能有很大影响.本文将基于能量的语音激活检测算法与基于模型的算法相结合来检测语音,在N IST2006核心测试数据集上,采用本文算法的系统相对于传统基于能量的方法性能最多有19%的提升.

Voice activity detection（VAD） is an algorithm to detect the voice endpoint.It can affect the performance of speaker recognition system greatly.In this paper,we combine the energy-based VAD method and the model-based VAD method to detect the voice endpoint.On the NIST 2006 SRE corpus,the proposed VAD algorithm can obtain 19% EER reduction over the traditional energy-based system at most.