基于变步长LMS减噪的基音检测改进算法

Improved algorithm for pitch detection based on variable step size LMS noise reduction

针对噪声环境下传统的基音检测算法精度不高的问题,提出一种改进的基音检测算法。采用变步长最小均方(LMS)自适应滤波器对带噪语音信号进行减噪,计算减噪后语音的自相关函数(ACF)归一化值、改进平均幅度差函数(MAMDF)、倒谱,建立非线性组合函数,突出基音周期处的峰值。基于置信区间和相邻语音帧的基频差值进行平滑处理,减少基音提取的错误。仿真结果表明,与传统算法相比,该算法检测准确率提高了至少6.7%,在低信噪比环境下鲁棒性也明显提高。

For low accuracy issues of traditional pitch detection algorithms in a noisy environment,an improved pitch detection algorithm was proposed.The variable step size least mean square(LMS)adaptive filter was used to reduce background noise.The normalized value of the auto correlation function(ACF),improved average magnitude difference function(MAMDF)and cepstrum were calculated and a nonlinear combining function was established to emphasize the peak of the true pitch period.Data smoothing processing was performed by calculating the confidence interval and the pitch frequency difference value between adjacent speech frames to reduce the pitch deviation.Experimental results show that,the proposed method can improve the accuracy of pitch detection by at least 6.7%compared with traditional algorithms.Moreover,the robustness is also improved significantly in low signal to noise ratio(SNR).