摘要
以小波变换及多分辨分析为理论基础,对语音端点检测中小波系数方差算法和子带平均能量算法进行了分析和研究,利用语音和噪声的频域差别,对这两种算法进行了优化,并应用于端点检测系统中,有效地改善了小波系数方差算法耗时长、实时性差的缺点,并克服了子带平均能量算法只对高斯白噪声检测效果好的局限性,提高了语音端点检测系统的实用性.通过MATLAB软件仿真的实验结果表明,采用优化算法的系统实现了语音端点检测准确性和快速性的最佳匹配,达到了此类检测设备的实用要求.
This paper studies the algorithm of sub-band average-energy and the variance of the wavelet coefficients in speech endpoint detection, which were on the basis of the principle of Wavelet transform and Muhi-resolution Analysis. It also optimizes the two algorithms. The proposed algorithm applied to the endpoint detection system improves the drawback of former algorithm in Long-Running and real-time badness which makes use of the variance of speech and noise in frequency characteristics. It also overcomes the limitation of only gaussian white noise is detected in the latter algorithm and improves applicability in speech endpoint detection. The simulation in the MAT- LAB software shows that system used optimization algorithm achieves optimal accurate and rapid matching. It meets the practical requirements of this kind of detecting equipment.
出处
《哈尔滨理工大学学报》
CAS
北大核心
2009年第1期51-54,59,共5页
Journal of Harbin University of Science and Technology
关键词
端点检测
小波变换
系数方差
子带平均能量
endpoint detection
wavelet transform
parameter variance
sub-band average-energy