Aplicação do filtro de Kalman nos coeficientes de aproximação e soft thresholding nos coeficientes de detalhe para a redução de ruído em sinais de voz no domínio Wavelet

Abstract

Noise is increasingly present in the transmission of speech signals, so new research is being developed to reduce it. This work presents a noise reduction technique that applies Kalman filtering to the approximation coefficients and soft thresholding to the detail coefficients of the wavelet decomposition speech signals contaminated by white noise. In order to evaluate the performance of this method, we measured the signal-to-noise ratio and the Itakura-Saito distance. We verified that the proposed technique showed higher output signal-to-noise ratios for input signal-to-noise ratios above 3dB and a lower spectral distortion when compared to the algorithm of Dhivya and Justin (2014).