Melhoria de sinais de fala baseada em atenuação espectral não-linear utilizando propriedades da prercepção auditiva

Abstract

Abstract: This paper aims to present an improvement of speech signals system, single channel, to be able to improve the quality and intelligibility of speech signals degraded by additive noise. The proposed algorithm is based on noise suppression rule of Ephraim and Malah, but with modifications made to handle noisy signals of low signal to noise ratio (SNR <10 dB). The main modification was made by introducing the concept of noise masking threshold, which is a well known property of the human auditory system. This FOMA, the solution became correlated with human perception. These changes were made to the original system of Ephraim and Malah in order to increase the noise reduction and simultaneously allow a further reduction of musical noise phenomenon. Experimental results showed that a general improvement in the quality of the speech was obtained for all types of approximately stationary noise considered in the experiment, in a wide range of signal to noise ratio.