Multiple Sound Source Localization by Using Diffuseness Estimation

This paper proposes a scheme for multiple source localization by using diffuseness estimation. According to the sparsity of the voice signal, single source time-frequency point (SSTP) always exists in the microphone recorded sound signal. The DOA estimates of these SSTPs have favorable localization statistical significance. Through analysis the characteristics of diffuseness measure, this paper gives an assumption that there is some correlation existed between diffuseness measure and SSTP. Through experiments, we found that the TF point with low diffuseness measure can be classified as SSTP. Therefore, a SSTP detection method based on diffuseness estimation is proposed. Meanwhile, the multiple source localization problem is converted to a single source localization among the detected SSTPs. The objective evaluations revealed that the proposed method achieved higher localization accuracy compared to some existing technique based on single sound source zone detection.