基于相似度的高精度基音检测算法

A similarity-based high resolution pitch detection algorithm

提出了一种具有较高精度且抗噪性能强的基音检测算法。该算法将线性预测残差看作语音源信号的近似,对其进行频谱分析,依据残差幅度谱算得基音周期的粗估值。然后回到时域信号,根据基音周期粗估值设计一长度可调的窗,通过窗函数在语音段连续取两段语音信号作相似度运算,可根据最大相似度值计算出准确的基音周期。该方法准确性高,在噪声环境下也具有较好的效果。

A pitch detection algorithm of high resolution and robust anti-noise is proposed in this paper. Firstly, the linear predictive residual is used as the approximate of original speech signal and is transformed by FFT. Secondly pitch is calculated based on the magnitude spectrum of residual and a ~ndow, the length of which could be adjusted, is designed. Finally the more exact pitch is obtained by similarity calculation of two successive speech signal segments chosen by the window in each frame signal. This algorithm is of high resolution and effective in low SNR environment.