ISCA Archive Interspeech 2014
ISCA Archive Interspeech 2014

Weighted spatial bispectrum correlation matrix for DOA estimation in the presence of interferences

Wei Xue, Shan Liang, Wenju Liu

Besides the undirected environmental noise, the surrounding interference also brings great challenges to the robust DOA estimation of the speech source. As conventional DOA estimation methods always assume an undirected noise model, they usually cannot perform reliably when the strong inference exists. In this paper, we propose a novel interference robust DOA estimation method, which is based on the “weighted spatial bispectrum correlation matrix (WSBCM)”. The WSBCM contains the spatial correlation information of bispectrum phase difference (BPD), and a new DOA estimator is further derived based on the eigenvalue analysis of the WSBCM. By formulating with WSBCM, the proposed method benefits from the redundant DOA-related information provided by the BPD. In addition, the WSBCM enables bispectrum weighting to highlight the pure speech units in the bispectrum, which further helps to improve the performance. In order to compute the bispectrum weights, a decision-directed approach is derived. The effectiveness of the proposed method is demonstrated by experiments conducted under various kinds of interference-existing scenarios.