Limitations of the spectral subtraction method on noise reduction

Abstract

Noise reduction has become very practicable with the development of sophisticated special purpose microcomputers and signal processors. As a result, many applications, such as mobile communication, speech recognition, military voice communications, broadcast link enhancement and restoration of musical recordings, very efficient noise reduction techniques are needed. A useful noise reduction algorithm is the spectral subtraction method which under some restrictions can extract an improved facsimile of the original signal by computing the spectrum of the noise and subtracting it from the combined corrupted signal spectrum. In this paper, the predicted performance of noise reduction using the spectral subtraction technique was investigated. During the study, signals were corrupted with channel noise derived from various sources and at different signal to noise ratios. The technique was applied to reduce the noise and found to be effective although the improvement of signal to noise ratio was found to be limited to 15 dB