基于语音信号稀疏性的FDICA初始化和后处理方法

FDICA Initialization and Post-Processing Method Based on Sparseness of Speech

目前解决语音信号盲源分离(Blind source separation,BSS)的两大类方法分别为频域独立成分分析(Frequency domain independent component analysis,FDICA)和基于稀疏性的时频掩蔽(Time frequency masking,TFmasking)。为此将两类方法优点相结合,利用TFmasking方法的结果,对FDICA做初始化,在加快FDICA收敛速度的同时也避免了次序不确定性问题。此外还提出了一种新的基于语音稀疏性FDICA的BSS后处理方法;基于局部最小比例控制(Local minimum ratio contr01led,LMRC)谱减法,比常规的TF masking、雏纳滤波等后处理方法,能够更有效地控制音乐噪声,提高分离性能。合成数据和实际采集数据的实验结果验证了所提方法的有效性。

There are two approaches being widely studied and employed to solve the blind source separation （BSS） problem. One is based on independent component analysis （ICA） and the other relies on the sparseness of source signals time frequency masking （TF-masking）. To speed up the convergence rate and to avoid permutation problems, a method combining the advantages of both methods is presented by using the results of TF masking to initialize the fre- quency domain ICA （FDICA）. Moreover, a new post-processing method for FDICA is pro- posed, i.e. local minimum ratio control （LMRC） spectral subtraction. It is based on the sparse characteristics of speech. Compared with the conventional TF masking and Wiener filter post processing methods, the proposed method can control musical noise more effectively, and improve the separation performance. Experimental results with synthetic data and real data demonstrate the effectiveness of the proposed method.