Transient Interference Suppression in Speech Signals Based on the OM-LSA Algorithm

Abstract:
Typical transient interferences, e.g. door knocks and keyboard tapping, are short in time, widely spread across the frequency domain, and have an abrupt nature. Thus, traditional speech enhancement techniques that use temporal smoothing to estimate the power spectral density (PSD) of the interference are inadequate. In this paper, we present a speech enhancement algorithm that suppresses transient interferences and pseudo-stationary background noise. The algorithm comprises an estimation of the transient and the pseudostationary noise PSDs, and enhancement of speech. The proposed algorithm is capable of tracking rapid variations of the input signal spectra and enables to effectively estimate the PSD of the transients. Experimental results show that the proposed algorithm is robust, does not rely on transient periodicity or reoccurrence, and exhibits good performance for various transient interference types. Index Terms— Speech enhancement, speech processing, acoustic noise, impulse noise, transient noise.