REAL-TIME SPEECH DERVERBERATION METHOD AND APPARATUS USING MULTI-CHANNEL LINEAR PREDICTION WITH ESTIMATION OF EARLY SPEECH PSD FOR DISTANT SPEECH RECOGNITION

摘要

Disclosed are a method and apparatus for real-time reverberation cancellation based on multi-channel linear prediction using early component estimation of voice power for far speech recognition. In a real-time reverberation cancellation method based on multi-channel linear prediction, a microphone input power corresponding to an input signal input through a plurality of channels and including an audio and reverberation signal and a voice power early path component (direct path & early reflection component) And estimating an early component of the voice power based on a cross component representing a correlation between late reverberation components of voice power, and presuming the estimated voice power with respect to the input signal. Removing reverberation based on the components.