Fault Detection in a Microphone Array by Intercorrelation of Features in Voice Activity Detection

Voice-based interaction generates a lot of interest for natural human–robot interaction. Auditory information from microphones is an essential clue for a robot's attention to a person. Fault detection of microphones is required in order to improve the reliability of the voice-based human–robot interaction. This paper proposes a new approach for real-time fault detection in a microphone array in conversation without a calibration signal and a known sound source position by the intercorrelation of features in voice activity detection. The approach is successfully applied to a six-microphone system, and experimental results show an average fault detection of 97.6%.