小波域降噪方法及其在历史音频保护中的应用

提出基于小波包变换的自适应多阈值音频降噪法,从小波包最佳分解树的各个终端结点选择不同的最佳阈值。而最佳阈值的选取是基于贝叶斯理论,并且采用折中阈值函数。实验和应用表明,相比小波包全局阈值降噪算法,此种降噪方法能使降噪后的音频保留尽可能多的原始音乐信号,较好地适合听众的听觉效应,满足历史音频资料数字化保存的需要。

Signal Denoising in Wavelet Domain and Its Application in Digital Preservation of Historical Audio Materials

Audio denoising is the preliminary step in the digital processing of historical audio materials.Wavelet domain denoising is an optional method.But traditional threshold shrinkage method has the tendency to over-suppress the wavelet coefficients in eliminating the HISS noise.This tendency may blur audio frequency and ruin the timbre.In this paper,an adaptive data-driven multi-threshold audio denoising method based on wavelet packet transform is proposed.In this method,the optimal threshold value at each terminal node of the wavelet packet tree with best decomposition is derived in a Bayesian framework,and the mean threshold function is presented.The experiments and applications show that in comparison with the formula of hard threshold,this method can retain more original music signals after HISS noise reduction,which adapts the digital music to the auditory response of listeners preferably.Thus,this method can meet the requirements of digital reservation of the historical audio materials in both subjective and objective aspects.