Lightweight and optimized sound source localization and tracking methods for open and closed microphone array configurations

摘要: Abstract Human–robot interaction in natural settings requires filtering out the different sources of sounds from environment. Such ability usually involves use microphone arrays to localize, track and separate sound online. Multi-microphone signal processing techniques can improve robustness noise but cost increases with number microphones used, limiting response time widespread on types mobile robots. Since source localization methods are most expensive terms computing resources as they involve scanning a large 3D space, minimizing amount computations required would facilitate their implementation The robot’s shape also brings constraints array geometry configurations. In addition, return noisy features that need be smoothed filtered by tracking sources. This paper presents novel method, called SRP-PHAT-HSDA, scans space coarse fine resolution grids reduce memory lookups. A directivity model is used directions scan ignore non significant pairs microphones. configuration method introduced automatically set parameters normally empirically tuned according array. For tracking, this modified Kalman (M3K) capable simultaneously Using 16-microphone low hardware, results show SRP-PHAT-HSDA M3K perform at least well other while using up 4 30 times less respectively.