基于稀疏表示和特征加权的离格双耳声源定位*

基于头相关传递函数数据库的传统双耳声源定位方法的定位角度往往被限定在头相关传递函数数据库的离散测量点上。当头相关传递函数数据库的测量方位角间隔较大时,这类算法的性能会显著下降,这就是典型的离格问题。该文提出了基于加权宽带稀疏贝叶斯学习的离格双耳声源定位算法。首先该算法建立离格双耳信号的稀疏表示模型,然后利用双耳相干与扩散能量比特征对各个频点进行加权以降低噪声和混响的影响,最后通过加权宽带稀疏贝叶斯学习方法估计离格声源的方位角。实验结果表明,该算法在各种复杂的声学环境下都有着较高的定位精度和鲁棒性,特别是提高了离格条件下的声源定位性能。

Off-grid binaural sound source localization using sparse representation and feature weighting

Traditional binaural sound source localization(BSSL)techniques using measured head-related transfer function (HRTF) databases often suffer a typical off-grid problem, where their estimated azimuth angles are restricted at the measured azimuth angles of HRTF databases. When the interval of the measured azimuth angles is large, the performance of these techniques will degrade significantly. This paper proposes an off-grid BSSL algorithm based on weighted wideband sparse Bayesian learning. First, the algorithm establishes an off-grid sparse representation model. Then weighted values based on binaural coherent-to-diffuse power ratio (BCDR) for each frequency band are calculated to reduce the impact of noise and reverberation. Finally, a weighted wideband sparse Bayesian learning method is derived to solve the off-grid BSSL problem. Experimental results show that the proposed method can achieve higher localization accuracy and is more robust than the compared BSSL techniques in various acoustic environments, especially under the off-grid situations.