Multi-source DOA estimation in reverberant environments using potential single-source points enhancement

Abstract Direction-of-arrival (DOA) estimation of multiple speech sources in enclosed environments is a hot topic in the acoustic field. Reverberation and multi-source overlapping are the primary challenges in multi-source localization. These disturbances result a performance degradation. In this paper, a new method for multi-source DOA estimation in reverberant environments is presented using potential single-source points enhancement. First, we made a statistical investigation about the effect of reverberation on single-source points (SSPs) whose components dominantly derive from only one source. From the experimental results, it is found that several time-frequency points detected as SSPs in anechoic environments are converted into non-SSPs when the signals are recorded in reverberant conditions. These points are defined as potential SSPs in this paper. Then, potential SSPs enhancement algorithm is designed to suppress the effect of reverberation for SSPs detection. Finally, DOA estimation is accomplished by post-processing the histogram, which is generated by the DOA estimates calculated at the detected SSPs. We conducted objective evaluations in both simulated and real environments. Experiment results indicate that the localization scheme proposed in this paper is more efficient with some state-of-the-art localization methods in terms of both DOA estimation and source counting.