基于EMD的基音检测预处理技术

针对基音检测的性能严重受背景噪声影响的问题,论文基于经验模态分解(EMD)理论,研究了含噪语音信号的EMD分解特性,参照小波阈值去噪方法,提出了一种基于EMD的自适应语音去噪算法,并且针对软、硬阈值函数的不足提出了一种新的阈值函数.MATLAB仿真结果表明,该方法可以有效地去除噪声,较好地恢复语音信号,与小波阈值去噪方法相比,信噪比、均方根误差等性能指标均有明显提高.

Technology of pitch detection pretreatment based on EMD

To solve the problem that pitch detection performance is seriously affected by the background noise, this paper researches the EMD characteristics of the noised speech based on empirical mode decomposition （EMD） theory, and introduces a self-adaptive speech signal denoising method referencing to the method of wavelet threshold. It also presents a new threshold function aiming at the disadvantages of soft and hard threshold function. According to Matlab simulation results, this algorithm could reduce the white noise effectively, and the performance indexes such as signal to noise ratio（SNR） and root mean square error （RMSE） both have improved significantly.