基于小波变换的语音基音周期鲁棒性检测  

A Robust Method For Speech Signals Pitch Extraction Based on Wavelet Transform

作　　者：樊桂花[1] 唐斌[1] 

机构地区：[1]解放军装备指挥技术学院测量控制系,北京101416

出　　处：《信号处理》2004年第4期365-368,共4页Journal of Signal Processing

摘　　要：固定门限的小波变换基音检测算法,在浊音信号起始段和结束段的模极大值不能有效提取。文中提出一种基于小波变换模极大值的基音周期自适应前后向跟踪算法,并结合过零率检测,实现了弱浊音信号基音周期的鲁棒性检测。算法对浊音信号采用自适应帧长,对陡变的基音周期能够准确提取,对噪声有较强的鲁棒性。A single threshold wavelet transform can't extract the pitch period at the beginning and the ending of sonant. Combining adjustable forward and backward pitch extraction with the average zero-crossing rate, the pitch period of weak sonant can be robust detected. With adapting frame, a sharply changed pitch can be detected exactly. The algorithm is less complicated and is suitable for real-time speech signals processing. Experimental results show that the algorithm has a better robustness.

关 键 词：语音信号 语音基音周期 鲁棒性检测 小波变换 

分 类 号：TN912.3[电子电信—通信与信息系统]