APPARATUS AND METHOD FOR FEATURE COMPENSATION FOR SPEECH RECOGNITION IN NOISE ENVIROMENT

摘要

The present invention relates to a feature compensation apparatus and method for speech recognition in a noisy environment. The feature compensation apparatus for speech recognition in a noisy environment according to the present invention comprises: a feature extraction unit which extracts voice signal feature information from a contaminated voice signal to which noise composed of two or more frames has been added; a noise estimation unit which estimates noise feature information based on the extracted voice signal feature information and compensated voice features; a probability calculation unit which calculates a correlation between frames adjacent to the contaminated voice signal to which the noise composed of the two or more frames has been added; and a voice feature compensation unit which generates the compensated voice features by removing noise features from the extracted voice signal feature information by taking into account the correlation between the frames adjacent to the contaminated voice signal and the estimated noise feature information.