A bioinspired spectro-temporal domain for sound denoising

CESAR ERNESTO MARTÍNEZ; JOHN GODDARD CLOSE; LEANDRO EZEQUIEL DI PERSIA; DIEGO HUMBERTO MILONE; HUGO LEONARDO RUFINER

Ciudad Autónoma de Buenos Aires

Simposio; XVII Argentine Symposium on Artificial Intelligence (ASAI 2016); 2016

SADIO Sociedad Argentina de Informática

The representation of sound signals at the cochlea and auditorycortical level has been studied as an alternative to classical analysismethods. In this work, we put forward a recently proposed featureextraction method called approximate auditory cortical representation,based on an approximation to the statistics of discharge patterns at theprimary auditory cortex. The approach here proposed estimates a nonnegative sparse coding with a combined dictionary of atoms calculatedfrom clean signal and noise. The denoising is carried out on noisy signals by the reconstruction of the signal discarding the atoms corresponding to the noise. Results on synthetic and real data show that the proposed method improves the quality of the signals, mainly under severe degradation. This communication corresponds to a journal paper published in 2015 in DSP (Elsevier).