摘要
提出了一种基于小波变换和平均周期的鲁棒性基音周期检测方法 ( MPRE)。检测前加入小波变换预处理 ,以平滑信号中的细节分量 ,检测采用最小预测能量差和基于子帧划分的基音平均周期检测方法。该方法可以克服很多基音检测方法中阈值难以确定的问题 ,并对噪声有较强的鲁棒性。
In previous pitch detection algorithms, such as AMDF algorithms, many decisions and thresholds are needed, but the thresholds vary with different environmental noises and speakers, which may greatly affect robustness of algorithm. A robust pitch detection algorithm based on wavelet-transformation and average pitch is proposed. The algorithm needs no thresholds, so it has great immunity to the effect of environmental noises and speakers. Before detection, a pre-process based on wavelet-transformation is performed to smooth the original speech signals, it further enhances the immunity to noise. In the pitch detection process, the frame is divided into several sub-frames, thus the average pitch of the frame can be detected. Experimental results show that the algorithm has a better robustness and is faster than the AMDF algorithm.
出处
《数据采集与处理》
CSCD
2002年第1期10-14,共5页
Journal of Data Acquisition and Processing
