Evaluation of different microphone arrays and localization algorithms in the context of ambient assisted living
This paper presents the evaluation of full 3D sound source localization systems for real world living environments.We tested several well-established algorithms. The Generalized Cross Correlation Phase Transform (GCC-PHAT) and Adaptive Eigenvalue Decomposition Phase Transform (AED-PHAT) algorithms were implemented as Time Delay of Arrival (TDOA) estimators. For the localization itself a three dimensional acoustic map was computed using the Global Coherence Field (GCF) as well as the modified Least Squares (LS) algorithm called Least Median of Squares (LMS). The combinations of these techniques are applied to different microphone array configurations composed of two fundamental microphone arrays. These fundamental arrays are a set of ceiling-mounted sensors with large distances and a small spherical array. Finally, a Voice Activity Detector (VAD) was applied in order to avoid false localization estimations during speech pauses. For evaluation we recorded a database of speech signals in a natural living environment. The results show that the combination of ambient microphone arrays with modern localization algorithms are able to locate people in a room in all three dimensions. However, the localization is not perfectly accurate and an error up to 0.4 m has to be tolerated.