A real time spectral subtraction based speech enhancement scheme

Various speech enhancement schemes are analyzed in terms of the conflicting real time requirements of computational delay, robustness and accuracy. A spectral subtraction scheme is found to be implementable in real time using the available digital processing board. The tasks such as the computation of the spectral contents of the speech plus noise and the noise, smoothing the estimates, and speech enhancement filtering are implemented in the frequency domain using FFT in view of the computational speed and robustness. A variant of a spectral subtraction scheme is implemented in real time on a DSP board and its performance is evaluated.