Improving real-time voice activity detection for perceptual robotic control in noisy environment

In order to enable natural human-robot interaction, robots will have to acquire the skills to detect and meaningfully integrate information from multiple modalities. In this paper, we focus on efficient voice activity detection (VAD) in the context of a multi-sensory robot device that uses audio activity information to drive perceptual function to yield natural and intuitive response to human behavior for robotic control. We proposed an improving VAD algorithm using frequency analysis, including harmonic spectral peaks (HSP), spectral flatness (SF) and spectral entropy (SE). The proposed method was compared the performance and advantages with five common VAD algorithms in three noise environments. We also described the way to integrate in our robot, Robert, and briefly introduced the perceptual multimodal framework for robotic control, where visual-audio modalities can be easily integrated such as speech content recognition and visual object identification or tracking.