低信噪比环境下改进的语音端点检测算法  

Improved Speech Endpoint Detection under Low SNR Environment

作　　者：董胡[1] 

机构地区：[1]长沙师范学院电子与信息工程系,湖南长沙410100

出　　处：《计算机技术与发展》2016年第3期71-74,共4页Computer Technology and Development

基　　金：湖南省科技计划项目(2012FJ3010);湖南省教育科研项目(12C0952);长沙师范学院科研基金项目(XXZD201218);长沙师范学院大学生研究性学习和创新性实验计划项目(DXYC201510)

摘　　要：端点检测在语音识别中具有非常重要的作用,其准确性将直接影响语音识别系统的正确率。为了提高低信噪比环境下语音端点检测的正确率,提出了一种基于多窗谱估计的改进谱减法和能量谱熵的端点检测算法。该算法首先利用多窗谱估计改进谱减法对含噪语音进行去噪以提高语音信号信噪比,接着对去噪后的语音信号使用新的能量谱熵算法进行端点检测。仿真实验结果表明,同常见端点检测算法相比较,该算法在低信噪比环境下具有较好的端点检测正确率且有一定的鲁棒性,证明了该算法的有效性。Endpoint detection has a very important role in speech recognition,its accuracy will directly affect the accuracy of speech recognition system. In order to improve the accuracy of speech endpoint detection under low SNR environment, an endpoint detection algorithm based on spectral subtraction of multitaper spectrum estimation and spectral entropy is proposed. Firstly,it uses improved spectral subtraction of multitaper spectrum estimation to denoise speech signal in order to improve signal to noise ratio, and then it utilizes energy-entropy-ratio algorithm to make endpoint detection for speech signal denoised, Simulation experiment results show that compared with common endpoint detection algorithm,this algorithm has good endpoint detection accuracy and certain robustness in low SNR environment. It proves the effectiveness of the proposed algorithm.

关 键 词：多窗谱估计 改进谱减法 谱熵 语音增强 端点检测 

分 类 号：TN912.35[电子电信—通信与信息系统]