A robust target signal detector based on statistical models using binaural cross-similarity information

This paper represents a robust target signal detector using two microphones. Most of target signal detection (TSD) algorithms incorporate with microphone arrays (MAs), and their basic assumption is that the location of target acoustic source is known. Therefore, target signal portions can be identified by exploiting the spatial information of the target location which is inherent in the binaural input signals. The proposed TSD employs a target signal absence probability (TSAP) estimation method that is statistically derived from the likelihood ratio test (LRT). A series of tests are conducted to evaluate the performance by comparing with a typical TSD method based on signal-to-interference-plus-noise ratios (SINRs). The experimental results show that the proposed method significantly outperforms the conventional TSD method in low SINR and vehicular noise environments.