The addition of environmental background noise to speech signals can be a serious problem in speech processing. It can make tasks such as voice recognition substantially more difficult and also significantly reduces the intelligibilty obtained with low bit-rate speech coding techniques (ref 1). The technique of spectral subtraction has been shown to be useful in reducing the level of narrowband stationary noise from speech (ref 2). This paper outlines the implementation and performance of an FFT based spectral subtraction noise reduction algorithm using a single TMS 32020 digital signal processing chip.