Single Channel Enhancement of Noisy Speech

The problem of enhancement of speech degraded by additive background noise has received considerable attention over the past decade. The reasons being, its wide range of applications and limitations of the available methods. Among the available single channel enhancement techniques, the spectral subtraction technique has been one of the widely adopted methods for suppressing stationary additive background noise. The greatest asset of spectral subtraction lies in its simplicity since all that is required is an estimate of the mean noise power. The basic idea of spectral subtraction is to obtain an estimate of the speech spectral level by subtracting the noise estimation from noisy speech. The spectral subtraction technique performs well as a pre-processor noise reduction technique for digital voice processors. In our work we first discuss the need for speech enhancement, its applications and the available different approaches. It is followed by classification of single channel enhancement techniques and brief overview of each method is given. The basic spectral subtraction method and a modified version that minimizes the shortcomings of the basic method have been discussed in detail. The criteria used for evaluating the performance of a speech enhancement method have been explained and conclusions are drawn on the performance of spectral subtraction method. Finally, the limitations of the present spectral subtraction method have been identified and suggestions for further improvement of the method have been put forward.