Speech Enhancement based on Sequential Noise Estimation with a Masking Property

Seiji Hayashi and Masahiro Suguimoto


Speech enhancement, Spectral subtraction, Noise estimate, Masking property


In this paper we describe a method for enhancing speech that is corrupted by additive background noise varying in time. The proposed nonlinear spectral-subtraction approach is based on sequentially updating the estimated noise per frame and adapting a masking property of the human ear. Furthermore, we developed an adaptive control of a scaling function, calculated by the regression line using the noise-masking signal-to-noise ratio (NMSNR). The proposed approach can efficiently remove additive noise related to various types of noise corruption.

Important Links:

Go Back