SPEECH CODING AND NOISE REDUCTION USING ICA­BASED SPEECH FEATURES

Jong­Hwan Lee 1 , Ho­Young Jung 2 , Te­Won Lee 3 , Soo­Young Lee 1, 2 Electronics and Telecommunications Research Institute
(TEL: +82­42­869­8031, FAX: +82­42­869­8570, E­mail: jhlee@neuron.kaist.ac.kr)

In this paper, we have obtained efficient speech fea­ tures using independent component analysis to human speeches. When independent component analysis is ap­ plied to speech signals for efficient encoding the adapted basis vectors resemble Gabor­like features. Then only a few active coefficients of the trained basis vectors are sufficient for encoding the speech signals. Those trained speech features can be used in automatic speech recognition systems, and the proposed method gives better recognition rates than conventional mel­frequency cepstral coefficients (MFCCs) features. Trained basis vectors can be also applied for the removal of Gaussian noise. Speech signal corrupted by additive white Gaus­ sian noise shows much improvements on the signal­to­ noise ratio (SNR) after the denoising process. Then, these denoised speech features show better recognition performances than MFCCs features.