论文部分内容阅读
针对传统自相关(ACF)基音周期检测算法存在较多的倍频和半频错误,文章提出一种基于线性预测残差域加权ACF基音周期检测方法。首先对语音信号中心削波,减小共振峰的影响;而后进行线性预测分析获得残差信号,对其求自相关值和循环幅度差(CAMDF)值,以CAMDF的倒数值为权重加权ACF进行基音周期检测;最后通过基音平滑算法对提取的基音轨迹进行后处理。仿真实验表明,该算法可降低基音提取的倍频和半频错误,提高估计精度。
Aiming at the existence of more frequency doubling and half-frequency errors in traditional auto-correlation (ACF) pitch detection algorithm, this paper proposes a method based on linear prediction residual-weighted ACF pitch detection. Firstly, the center of the speech signal is clipped to reduce the effect of the formant. Then, a linear prediction analysis is performed to obtain the residual signal, and the autocorrelation value and the difference of the cyclic amplitude (CAMDF) are obtained. The weight of the CAMDF is used to weight the ACF Pitch detection; Finally, the pitch-smoothing algorithm is used to post-process the extracted pitch track. Simulation results show that this algorithm can reduce the frequency doubling and half-frequency errors of pitch extraction and improve the estimation accuracy.