基于BP神经网络的双耳声源定位算法

Binaural sound source localization based on BP neural network

为了提高复杂声学环境中双耳声源定位性能,本文利用多种双耳空间特征参数,提出了基于神经网络方法的鲁棒双耳声源定位算法。本文将不同声学环境下双耳声信号对应的互相关函数、耳间强度差等空间定位线索,作为输入特征,用于训练BP神经网络。在混响和噪声环境下,与传统双耳定位算法相比,本文基于BP神经网络的双耳定位算法的定位性能有显著提高,特别是低信噪比条件下,定位正确率提高更为显著。

In order to improve the performance of binaural sound source localization （SSL） in complex acoustic environ- ment, this paper proposes a robust binaural sound source localization algorithm based on BP neural network using binaural spatial cues. In this paper, localization cues of the binaural signals in different acoustic environments, such as the cross - correlation function and inter- aural intensity difference, are utilized as input features to train BP neural networks. Compared with traditional binaural SSL algorithm, the performance of the proposed algorithm has been significantly improved, especially under low SNR condition.