基于先验信噪比和能零熵的语音端点检测算法  

Endpoint Detection Algorithm with Priori SNR and Energy-zero-spectral Entropy

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》2017年第7期72-75,共4页Computer Technology and Development

基　　金：湖南省教育厅科学研究优秀青年项目(17B025);湖南省教育厅科学研究项目(12C0952);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);长沙师范学院院级科研项目(XYYB201517)

摘　　要：端点检测技术是语音识别系统中的一项关键技术,其性能在某种程度上对整个语音识别系统有着较大的影响,传统的语音端点检测算法在低信噪比环境下存在端点检测正确率低、抗噪性能差等问题。针对传统端点检测算法在低信噪比环境下存在的上述问题,提出了一种基于先验信噪比估计和能零熵的语音端点检测算法。该算法通过改进的先验信噪比估计算法对含噪语音进行增强处理,并对增强后的语音信号设置自适应端点检测阈值,利用能零熵算法对增强后的语音信号进行端点检测,实现了低信噪比环境下的语音端点检测。仿真实验结果表明,与传统的能零积和谱熵端点检测算法相比,所提出的端点检测算法在不同的低信噪比环境下具有较好的鲁棒性与较高的端点检测正确率。Speech endpoint detection is the key technology in voice recognition system, and its performance has a great influence for speech recognition system in some extent. However,traditional speech endpoint detection algorithms have problems of low accuracy and poor anti-noise under low SNR environment. In order to solve the problems above, a kind of speech endpoint detection algorithm based on prior SNR estimation and energy-zero-entropy has been proposed, in which the improved prior SNR estimation algorithm is employed to make speech enhancement processing of speech with noise and then the adaptive endpoint detection threshold is set for the enhanced speech signal. The energy-zero-spectral entropy algorithm is eventually adopted to make endpoint detection for enhanced speech signal and the speech endpoint detection under low noise environment is achieved. Simulation experiment results show that compared with tradi- tional energy-zero-product and spectrum entropy endpoint detection algorithm,the proposed endpoint detection algorithm has better ro- bustness and hi^her endpoint detection accuracy in different low SNR environment.

关 键 词：先验信噪比 鲁棒性 端点检测 谱熵 能零积 

分 类 号：TN912.35[电子电信—通信与信息系统]