基于模极大值和尺度理论的音频降噪方法

Audio denoising method based on wavelet transformmodulus maximum and scale tracking theory

在数字化时代,音频的转录或录制都会引入噪音,但是历史音频保存和音频资料处理需要纯净的音频信号,因此音频降噪研究有着重要的现实意义。首先介绍了二进小波和奇异性指数,并阐述了尺度跟踪和模极大值重构等理论,在M allat工作的基础上,提出了一种基于小波滤波的音频降噪方法。该方法首先引入补偿因子削减二进小波变换对系数造成的影响,并计算带噪音频的小波系数和模极大值;然后基于信号和噪声奇异指数不同的特点,结合阈值降噪和尺度跟踪理论,采用层间相关搜索去除噪声的模极大值;最后利用交替投影算法,重建音频信号。用该方法处理带c lick和h iss噪声的音频信号,跟小波阈值方法和小波包方法相比,能达到较好的听觉效果和信噪比。同时观察信号的波形图及模极大值演示图,发现该方法都表现出优异的降噪效果。

Audio recording or transcription inevitably brings in noise,so audio denoising is crucial for data processing and preservation.There are many existing techniques with respect to audio denoising based on wavelet transformation.An improved algorithm was provided based on three kinds of techniques: dyadic wavelet,scale tracking theory and modulus maximum theory.The novelties of the algorithm lie in the following: the compensation factors is introduced in order to reduce the influence of scale discretization;th...