Noise reduction using perceptual spectral change

This article deals with the problem of noise reduction for hands-free sound pick-up. Our context is more precisely the sound pick-up using single microphone in cars. In such environments, the speech is highly corrupted by engine and aerodynamics noises. The methods proposed here try to enhance noise reduction systems based on Wiener filtering in the frequency domain by taking into account the short time variations of speech. It aims in fact at adapting the estimation of the signal spectrum and the estimation of the noise spectrum to the spectrum variations of the noisy signal. By preserving voice attacks, distortion on the useful signal is reduced as shown in this article. Background residual noise and musical tone phenomenon are also discussed about.