ABSTRACT
This chapter describes a family of algorithms that are capable of improving speech intelligibility. The corruption of the fine structure and introduction of stochastic envelope fluctuations associated with the inaccurate estimates of the noise intensity and nonlinear processing of the mixture envelopes further diminished speech intelligibility. It was argued that it was the stochastic effects that prevented spectral subtractive algorithms from improving speech perception in noise. The frequency-specific gain function applied to the noisy speech spectrum is far from perfect as it depends on the estimated signal-to-noise ratio and estimated noise spectrum. Despite the fact that the gain function is typically bounded between zero and one, the target signal may be overamplified because the soft gain function is applied to the noisy speech spectrum. The direct consequence of applying binary gain functions to the noisy speech spectrum is that certain channels are retained while other channels are discarded.
