Antinoise and Antiairflow Sound Source Tracking System for Drones

Sound source localization on drones processes the audio signals collected by acoustic sensor arrays mounted on drones to obtain the sound source location. It has significant applications in military and disaster rescue operations. However, the substantial ego-noise of drones and the airflow generated by propellers around the drone greatly reduce the accuracy of localization. Additionally, in practical scenarios, the relative position between the drone and the sound source often changes over time, posing further challenges to localization. This paper implements an anti-noise and anti-airflow sound source tracking system on a drone. The system consists of two parts: the software part, which includes the anti-noise and anti-airflow sound source tracking algorithm, and the hardware part, which is composed of a custom drone verification platform. The algorithm has been validated on the DREGON dataset, achieving better localization performance compared to five other algorithms, including the state-of-the-art time–frequency spatial filtering (TFS), the state-of-the-art cross-correlation based direction of arrival (DOA) estimation technique, the generalized cross-correlation phase transform (GCC-PHAT) method, the steered response power deep neural network (SRP-DNN), and the state-of-the-art time–frequency spatial filtering deep neural network (TFS-DNN). In the proposed algorithm, different sub-segment lengths will affect the localization accuracy. Therefore, the influence of different sub-segment lengths was compared, revealing that the highest accuracy was attained when the sub-segment length was 1s. In addition, the algorithm is also validated on the custom drone verification platform—which consists of an F450 drone, a circular eight-microphone array, and a Sololinker development board.