SPEECH ENHANCEMENT APPARATUS USING CODEBOOK AND METHOD THEREOF

摘要

The present invention relates to a device and a method for improving a voice based on codebooks. A clean device based on codebooks comprises: a linear prediction coding (LPC) extraction unit to extract an LPC coefficient for a clean signal and a noise signal that are included in an inputted voice signal; a parameter generation unit to generate line spectral frequency (LSF) parameters corresponding to the extracted LPC; an LSF gain estimation unit to calculate at least one distortion value by combining clean LSF codebooks matched with the LSF parameters in prestored clean codebooks and noise LSF codebooks matched with the LSF parameters in prestored noise codebooks; a maximum likelihood estimation unit to detect respective indexes for code vectors of the clean signal and the noise signal for the distortion value based on a predetermined maximum likelihood (ML) estimation manner; a spectrum gain estimation unit to estimate gain values of the clean signal and the noise signal that are matched with the respective indexes in the prestored gain values for the clean signal and the noise signal; and a filter unit to filter the received signal based on the gain values of the clean and noise signals, thereby remarkably cutting a searching process compared to an existing manner of using only the ML or MMSE and remarkably reducing a load of a system due to the operation process.