System and method for noise robust speech recognition using histogram equalization with window-based test cumulative distribution function estimation

摘要

A voice recognition system and a method using histogram equalization technique by window base cumulative distribution function estimation are provided to reduce the degradation of voice recognition performance in the environment with noise. A voice input unit converts an analog voice signal into a digital signal(200). A characteristic extraction unit extracts an acoustic feature parameter in the frame unit(202), after dividing the digital voice signal in to a regular time interval(201). A characteristic compensation unit extracts a compensated feature parameter(203). A voice recognition engine performs a voice recognition and outputs the acknowledged result(204).