Algoritmo de redução de ruído em sinais de fala baseado em filtros estocásticos

Abstract

This undergraduate research project focuses on developing a noise reduction algorithm for speech signals based on stochastic filters, aiming to mitigate the interference of various types of noise that compromise the signal quality. Based in a literature review, algorithms are developed that employ Kalman Filters (KF) whose Autoregressive (AR) models are obtained through linear prediction that together with other digital signal processing techniques are dimensioned to model the complex behavior of speech. Additionally, the improvement of these algorithms is proposed through the implementation of the Interactive Multiple Models (IMM) filter. The results of this study indicate that the developed algorithms are effective in reducing Gaussian and colored noise, and highlight the ability of IMM-based algorithms to improve the accuracy and intelligibility of signals compared to KF-based algorithms.