Design and Analysis of Multi-Source Extraction and Localization for Sound-Activated Designation of Automata

In this paper, we utilize speech signal to design sound-activated designation for confining an automaton to specific physical space. This work will present the design and analysis of multi-source extraction and localization to application of ubiquitous sound activation system. To extract a speaker sound from interference sources (such as a babble noise generated from TV playing), a multi-source extraction (MSE) process is used to perform noisy speech separation using a Fast-ICA (independent component analysis) with subspace-based speech enhancement. After extracting the speaker signal, the next step is to perform the speaker localization by adopting the minimum variance distortionless response (MVDR) method which can be used to estimate the time difference of arrival (TDOA) of a speaker. The simulation process of multi-source extraction and localization makes use of the microphones mounted in the experimental environment of a 167′′ × 228′′ × 91′′ demo room with a two-element microphone array. We tested three noise signals including background noise with/without babble noise and robot engine noise. The experimental results show that the proposed designation system is promising for confining an automaton to specific physical space in a noisy environment.