摘要
本文探讨了基于时域的语音切分算法,在前人研究的基础上,提出一种改进算法——自适应、前后搜索和检测短时脉冲噪音算法。该算法主要利用语音信号的短时参数,采用统计的方法定出切分所需要的阈值;根据背景音和静音过零率的不同,进一步搜索符合要求的静音帧;同时滤去短时脉冲噪音。实验证明,该算法准确卑很高,有很好的鲁棒性,允许误差在60 ms 的范围内,对于原始语音切分错误率为5.04%;在信噪比(SNR)大于等于2 dB 的情况下,对带噪语音的切分错误率为10%～20%。
This paper researches on speech detection algodthrn based on time domain, and describes an adaptive, both forwards and backwards search, detecting short-term pulse noise algorithm. This algorithm uses a variety of features including the frame amplitude and zero crossing rate to calculate threshold using statistical method. And it searches much further for the unvoiced frame according to the ZCR（ zero crossing rate ）, which is differ from unvoiced frame to background frame. This algorithm also detects impulse noise that last little. Experimental results show that this im- provement has good performance, even in noisy condition. Testing the original speech, the error rate is 5.04%, and in noisy environment with a SNR of beyond 2 dB, the rate is around 80-90%.
出处
《计算机科学》
CSCD
北大核心
2006年第4期164-167,共4页
Computer Science
基金
国家自然重点基金资助
基金号:60433030
关键词
语音切分
短时参数
自适应
前后搜索
检测短时脉冲噪首
Speech detection, Short-term variety, Adaptive, Search forwards and backwards, Detecting short-term pulse noise
