VOICE ACTIVITY DETECTION USING VOCAL TRACT AREA INFORMATION

摘要

A voice activity detection (VAD) system includes: an input processing module receiving an acoustic signal, converting the acoustic signal into an analog signal, and then, converting the same into a digital signal; an energy-based detection module receiving one of the analog/digital signals, and determining a sound activity judgment; an area function-based detection module deriving an area-related function from the digital signal, and outputting an area-based judgment in accordance with the area-related function through a machine-learning method; and a VAD judgment module making a final VAD judgment based on the sound activity judgment from the energy-based detection module and the area-based judgment from the area function-based detection module. Therefore, the system can have low complexity to prevent unnecessary activation in a high-power mode.