论文部分内容阅读
人耳具有提取感兴趣的声音、屏蔽噪声以及定位的功能.利用人耳听觉机理的听觉场景分析技术是语音信号处理的有力工具.提出了一种新的基于听觉场景分析的盲分离方法,可以对多个源信号混合而成的2个传感器信号进行分离.耳间时间差(ITD)和强度差(IID)包含信号源的位置信息.通过听觉外周模型对混合信号进行处理,得到各个频率通道的ITD和IID,再对混合信号的频域表示进行聚类,可得到源信号的频域表示.由于ITD和IID分别在高低频段起主要作用,聚类时对高低频段采取了不同的分类准则.仿真实验表明了该方法的有效性.
The human ear has the function of extracting the sound of interest, shielding the noise and locating.The auditory scene analysis technology using human ear auditory mechanism is a powerful tool for speech signal processing.This paper proposes a new blind separation method based on auditory scene analysis, Separating the two sensor signals from a mixture of multiple source signals The ITD and IID contain the location information of the signal source.The mixed signal is processed by the auditory peripheral model to obtain the ITD and IID, and then cluster the frequency domain representation of the mixed signal to obtain the frequency domain representation of the source signal.Due to ITD and IID play a major role in high and low frequency bands, different classification criteria are adopted for high and low frequency bands during clustering. Simulation results show the effectiveness of this method.