Design and In-Field Testing of Target Sound-Source Positioning with Insights from Indoor Acoustic Environment

The sound-source localization method derived by combining the steered-response power phase transform (SRP-PHAT) method with stochastic region contraction (SRC) is one of the most effective localization methods currently available. It can yield accurate target sound-source localization results in weak-noise and moderate-reverberation environments. However, owing to the unstructured room space (which entails many points to be searched), the SRP-PHAT-SRC localization method using grid searches is computationally heavy and exhibits poor real-time performance. Therefore, we propose an improved method using insights from the indoor acoustic environment; our model estimates the room geometry via acoustic scene reconstruction and triangulates this estimated geometry via (offline) Delaunay triangulation for structured room space volumes [the volumes are selectively assigned (online) to SRC initialization when positioning]; the method searches for the target sound source within a more trusted space in the room and thereby eliminates computationally wasteful searches in regions where the sound source does not appear. When the position estimates of the motion target are updated, an increased number of points (position estimates) can be used to update the room space-volume structure (these are local updates that do not increase online positioning complexity), making the selective volume more efficient for the next searching task. We have verified the feasibility of the proposed method in improving positioning time consumption and demonstrated its good practical application prospects.