METHOD FOR PROCESSING SPEECH IN THE PRESENCE OF ACOUSTIC NOISE: METHOD OF NON - LINEAR SPECTRAL SUBTRACTION.

摘要

P To process a speech signal to attenuate the contribution of the noise before transmission, each frame of the sampled, digitized and formatted signal in frames of a given length is subjected to a processing of the form: BR/ (CF DRAWING IN BOPI ) BR/ BR/ applied to the transform of each frame in the frequency domain, where: BR/ - | X [omega] | is the estimate of the amplitude of the speech affected by noise; BR/ - | B [omega] | is an estimate of the amplitude of the noise, obtained from an observation of the noise, during silent periods of speech. | B (omega) | estimate using an estimator with a sufficient time constant; BR/ - alpha (omega) is a noise overestimation factor, calculated on a predetermined number of noise frames upstream of the processed frame; BR/ - opt [alpha (omega)] is a weighting of the subtractive term which is a function of the signal-to-noise ratio in a particular frequency band, this weighting being bounded. / P