A Perceptually Motivated Multi-Band Spectral Subtraction Algorithm for Enhancement of Degraded Speech

The spectral subtraction method is a classical approach for enhancement of degraded speech. The basic principle of this technique is to estimate the short-time spectral magnitude of speech by subtracting estimated noise from the noisy speech spectrum and to combine it with the phase of the noisy speech. Besides reducing the noise, this method generates an unnatural and unpleasant noise, called remnant noise. The other drawback of this method is that it can work only for white Gaussian noise which has a flat spectrum and is distributed uniformly over the frequency spectrum. But real-world noise is mostly colored and has a non-uniform spectrum. To take care of this kind of noises, spectral subtraction algorithm has been extended to a multi-band case with uniformly spaced frequency bands. In this paper, a perceptually motivated multi-band spectral subtraction algorithm is proposed to enhance the speech signal degraded by colored noise. In the proposed scheme, the whole speech spectrum is divided in different non-uniform bands (N = 6) in accordance to the critical-band rate scale and spectral subtraction is executed independently in each band. The simulation results as well as informal subjective evaluations show that the proposed algorithm reduces remnant noise efficiently and the enhanced speech contains minimal speech distortions with improved signal-to-noise ratio.