A speech spectral estimator using adaptive speech probability density function

In this paper, we propose a speech spectral enhancer based on the MAP estimation using variable speech probability density function (PDF). The proposed speech enhancement algorithm adaptively changes the speech PDF used in the MAP estimation according to the observed spectral power. In speech segments, the speech spectral density approaches a Rayleigh distribution to keep the quality of the enhanced speech. In non-speech segments, it approaches a delta function to reduce noise effectively. The proposed technique is effective in suppressing residual noise well. Computer simulation results show that the proposed speech enhancer is superior to the conventional methods in the noise reduction capability.