一种基于Matching pursuits时频分解算法的语音降噪方法

New Method to Canceling Noise in Speech with Matching Pursuits Algorithm in Time-Frequency Domain

提出一种基于时频分解算法的语音降噪方法,利用Matchingpursuits时频分解算法对含噪语音信号进行分解,把分解后的各个分量的魏格纳分布之和作为整个信号的魏格纳分布,这样便使得信号和噪声在时频平面上有了较直观的分布,利用这个特点对信号进行降噪。仿真结果表明,该方法能在保证可听懂度的情况下,有效地去除含噪语音信号中的宽带噪声。

In speech recongnition,the main goal of noise cancellation is to eliminate the noise as possible with good quality. With the study of the speech signal in the time-frequency domain, a new approach which based on Matching pursuit decomposition algorithm is proposed, by which to decompose the signal in time-frequency domain,then take the sum of Wigner distribution of each part signal as the whole signal Wigner distribution, thus an intuitionistic distribution of signal is shown on the time-frequency platform. Simulation results show that under the limitation of speech understandability, the new approach can eliminate the wide-band noise effectively.