A knowledge based real-time speech detector for microphone array videoconferencing system

This paper presents a novel speech detection method for a microphone array system that is able to effectively detect voice-only events and distinguish against nonspeech and speech-like sounds. The method is robust and suitable for a highly attenuated far-field recordings environment. Using a new 2-dimensional feature set derived from the knowledge of speech production and noise generation, the proposed method has shown its effectiveness in broad-class sound classification, which is observed in the real data simulation shown in this paper.