Optimización de Algoritmos de reducción de ruido en señales de voz.

Abstract

Abstract: The present Curricular Integration Project focuses on the analysis and optimization of classical algorithms for noise reduction in speech signals, whose application contributes to improving signal quality and preserving the useful information they contain in noisy environments. This research is supported by a theoretical framework based on digital communications and statistical signal processing, addressing the nature of noise in communication systems and its impact on speech intelligibility and perceptual quality. The mathematical modeling of Wiener filtering, Kalman filtering, spectral subtraction, and adaptive LMS/NLMS filtering algorithms was developed according to optimal estimation and mean square error criteria. Using a controlled experimental environment, the performance of each technique was evaluated through temporal and spectral analyses of the signals, employing objective parameters such as the signal-to-noise ratio (SNR) and magnitude spectrum analysis. The results show that adaptive methods provide the best trade-off between noise suppression and preservation of speech characteristics, highlighting their suitability for telecommunications and speech processing applications.