Evaluation of clipping-noise suppression of stationary-noisy speech based on spectral compensation

Development of communication systems allows people to easily record and distribute their speech. The clipping-noise, however, degrades the sound quality in the speech recording when gain level of input signals is excessive in the maximum range of an amplitude. In this case, it is necessary to suppress the clipping- noise in the observed speech for improving its sound quality. Although a linear prediction method has been conventionally proposed for suppressing the clipping-noise, it has a problem with degradation of the restoration performance by cumulating error when the speech includes a large amount of the clipping-noise. This paper describes a method for the clipping-noise suppression for the stationary-noisy speech based on the spectral compensation in a noisy environment. In this method, to suppress the clipping-noise, the Gaussian mixture models are utilized for modeling the power spectral envelope of the speech on each frame in the lower frequency band. The clean speech signals in a database are also utilized for restoring the clipping speech in the higher frequency band. We carried out evaluation experiments with a speech quality, and confirmed the effectiveness of the proposed method for the speech which includes a large amount of the clipping-noise.