基于欠定盲源分离的双路音频信号噪声自适应分离

Dual channel audio signal adaptive noise separation based on underdetermined blind source separation

当多个源信号同时存在于同一频段或时间域内时,它们可能会相互干扰,导致信号混叠。这种情况下使用双路音频传感器进行捕捉,无法准确地捕捉到所有源信号的信息,导致分离过程具有不确定性。对此,提出一种基于欠定盲源分离的双路音频信号噪声自适应分离方法。首先,构建欠定盲源分离模型,基于小波包变换分解和重构信号获取信号分量,并依据信号和分量之间的互相关系数筛选分解后的分量,删除其中的冗余分量后生成新的观测信号;然后,依据贝叶斯信息准则的奇异值分解方法估计该源信号的数量,将其转换为正定白化信号;最后,利用快速独立成分分析法将该信号分类,实现双路音频信号噪声自适应分离。测试结果显示:所提方法能够在保证信号质量的前提下完成信号变换处理,信干比均在15 dB以上;筛选后保留的各个分量相关系数均在0.65以上,有效地完成了对信号噪声的分离。

When multiple source signals coexist in the same frequency band or time domain,they may interfere with each other,resulting in signal aliasing.In this case,using dual channel audio sensors for capture cannot accurately capture the information of all source signals,resulting in uncertainty in the separation process.A dual channel audio signal noise adaptive separation method based on underdetermined blind source separation is proposed.An underdetermined blind source separation model is constructed,which can decompose and reconstruct the signal based on wavelet packet transform to obtain signal components.The decomposed components are selected based on the number of interrelationships between the signal and the components,and redundant components are removed to generate new observation signals.Based on the Bayesian information criterion,the singular value decomposition method is used to estimate the quantity of the source signal and convert it into the positive definite white signal.The fast independent component analysis method is used to classify the signal,so as to realize adaptive noise separation of dual channel audio signals.The testing results show that this method can complete signal transformation processing while ensuring signal quality,the signal-to-noise ratio results are all above 15 dB,and the correlation coefficients of each component retained after screening are all above 0.65,which can effectively separate signal and noise.