Proposal of Practical Sound Source Localization Method Using Histogram and Frequency Information of Spatial Spectrum for Drone Audition

Abstract

A technology to search for victims in disaster areas by localizing human-related sound sources, such as voices and emergency whistles, using a drone-embedded microphone array was researched. One of the challenges is the development of sound source localization methods. Such a sound-based search method requires a high resolution, a high tolerance for quickly changing dynamic ego-noise, a large search range, high real-time performance, and high versatility. In this paper, we propose a novel sound source localization method based on multiple signal classification for victim search using a drone-embedded microphone array to satisfy these requirements. In the proposed method, the ego-noise and target sound components are extracted using the histogram information of the three-dimensional spatial spectrum (azimuth, elevation, and frequency) at the current time, and they are separated using continuity. The direction of arrival of the target sound is estimated from the separated target sound component. Since this method is processed with only simple calculations and does not use previous information, all requirements can be satisfied simultaneously. Evaluation experiments using recorded sound in a real outdoor environment show that the localization performance of the proposed method was higher than that of the existing and previously proposed methods, indicating the usefulness of the proposed method.

Keywords

• drone
• drone audition
• search and rescue
• sound source localization
• microphone array
• multiple signal classification