摘要
传统的端点检测在信噪比较高的环境下可以有效地对语音进行端点检测,但是在低信噪比环境下端点检测的正确率急剧下降。针对在低信噪比环境下语音端点检测正确率不高的问题,提出一种将调制域谱减法和对数能量子带谱熵相结合的的端点检测算法;该算法首先利用调制域谱减法去除带噪语音的噪声以提高语音信号的信噪比,然后结合对数能量和子带谱熵算法对消噪后的语音信号进行端点检测。实验仿真结果表明,该算法在低信噪比环境下能有效提高语音端点检测的正确率且具有一定的稳健性。
For improving the accuracy rate of speech endpoint detection in low SNR environment, speech endpoint detection based on modulation domain spectral subtraction and improved spectral entropy is presented. Firstly the gathered speech signals s SNR was improved by modulation domain spectral subtraction algorithm. Then the logarithmic energy and band-partitioning spectral entropy algorithm to make endpoint detection for enhanced speech were used. The simulation results show that the method for speech endpoint detection in low SNR is effective and it has certain robustness.
出处
《科学技术与工程》
北大核心
2017年第21期50-56,共7页
Science Technology and Engineering
基金
国家自然科学基金(61461011)
教育部重点实验室2016年主任基金(CRKL160107)资助
关键词
低信噪比
调制域
谱减法
子带谱熵
对数能量
端点检测
low SNR modulation domain spectral subtraction band-partitioning spectral entropy logarithmic energy endpoint detection
