利用小波变换加权自相关的基音检测法

Weighted Autocorrelation Method for Pitch Detection Based on Wavelet Transform

自相关函数法、平均幅度差函数法及小波变换法是经典的基音检测方法,本文简要分析了单独使用它们进行基音检测时存在的不足,提出了一种基于小波变换的加权自相关的检测方法。将多级小波变换的近似分量加权求和以突出基音信息,采用改进的平均幅度差函数加权自相关函数的方法以突出真实基音周期处的峰值,提高基音检测的正确率。实验表明,与传统的自相关函数法和平均幅度差函数法相比,本文方法减少了倍频和半频错误,提高了基音检测的精度,在信噪比为-5 dB时仍能得到较准确的结果。

Autocorrelation function, average magnitude difference function and wavelet transform methods are the classical pitch detection methods. The shortcomings of these methods are analyzed when used alone. A weighted autocorrelation method is proposed for the pitch detection based on the wavelet transform. The approximate components of multi-level wavelet transform are weighted to emphasize the fundamental frequency information. Besides, to emphasize the peak at the position of the true pitch period, the method of autocorrelation function. weighted by the modified average magnitude difference function is used. Compared with the classical methods, the method can decrease double and half frequency errors and improve the pitch detection precision. Moreover, it can obtain accurate results when the signal to noise ratio is -5 dB.