Speech Signal Intelligibility Improvement by Statistical Signal Processing

Abstract

The speech signal is an effective carrier of information between communicating persons or in public addressing systems. A clean and clear speech signal efficiently conveys the information to the intended destinations. But the background reverberation and noise addition eventually lead to degradation of signal quality and some perceptual aspects such as intelligibility and degree of listener fatigue are affected to some level. It is required to decrease the noise content in the speech signal and improve its intelligibility. This paper articulates the statistical methods and techniques used in suppressing the background noise in a speech signal and improve its intelligibility by removing the unwanted portion of the signal. The statistical signal processing methods are based on observed data analysis and inferences rather than pure mathematical postulations and calculations.