基于子带信噪比估计和软判决的鲁棒双耳声源定位算法

Robust binaural sound source localization based on sub-band SNR estimation and soft decision

为了提高噪声和混响环境下的双耳声源定位算法性能,提出了一种基于子带信噪比估计和软判决的双耳互功率谱和耳间时间差估计算法.首先根据每帧中每个子带双耳声信号的自相关矩阵估计子带信噪比;其次,将子带信噪比映射为软判决值,并对双耳互功率谱进行加权;最后利用加权后的互功率谱估计耳间时间差,从而判断目标声源方位.仿真测试和实际环境测试均表明:与基于互相关函数、过零率的传统双耳声源定位算法相比,所提算法在噪声和混响的复杂声学环境下,显著提高了双耳声源定位性能.

In order to improve the localization performance in noisy and reverberation environments, a robust binaural sound source localization （SSL） algorithm based on sub-band signal-to-noise ratio （SNR） estimation and soft decision is proposed. First, sub-band SNR is estimated based on the au- tocorrelation matrix of sub-band binaural sound signals in each frame. Then, the sub-band SNR is mapped to soft decision value, and the cross power spectrum density （PSD） of binaural sound signal is weighted by soft decision. Finally, inter-aural time difference （ITD） is computed by weighted cross PSD, and the azimuth of sound source is estimated. Simulation and real environment test re- suits show that, compared with the conventional binaural SSL algorithms based on cross correlation and zeros crossing, the localization performance of the proposed algorithm is significantly improved in complex acoustic environments.