一种改进的自相关函数基音检测算法

提出了一种改进的自相关函数基音检测算法:首先利用清浊音自相关函数幅度值的不同性质进行清浊音判决,然后仅对浊音段检测基音周期。在基音检测前用带通滤波、中心削波和数值滤波等方法进行预处理,去除共振峰和高频噪声的影响;在基音检测过程的后端用搜索平滑方法进行后处理,平滑掉半、倍频点和随机错误点。实验结果表明,改进算法的效果优于传统自相关算法,而且在信噪比低至5 dB时仍有良好的清浊音判决和基音检测性能。

Improved Pitch Detection Algorithm Based on Autocorrelation Function

An improved pitch detection algorithm based on autocorrelation function is proposed.Unvoiced/voiced decision is realized using the differences between unvoiced and voiced autocorrelation function.Before the pitch detection,the voiced is pretreated by band-pass filter,center clipping and digital filter to reduce the effects of formant and high-frequency noises;after the pitch detection,searching smoothing method is exploited to overcoming the multiple or half frequency errors and random errors.The experimental results show that the performance is superior to traditional autocorrelation function-based algorithm.Furthermore,this new method still works well under low signal noise ratio(SNR=5 dB).