R. Thoonsaengngam and N. Tangsangiumvisai On Improvement
of the a priori SNR estimation via gain modification for speech enhancement
|
Abstract
This paper presents an improved approach for speech enhancement in single-microphone applications. The proposed Noise Suppression (NS) technique is based upon a modified a priori Signal-to-Noise Ratio (SNR) estimation. The proposed a priori SNR estimator aims at tracking very fast the rise and fall of the speech spectrum, while reducing the musical noise effect. This yields an efficient NS technique that provides minimum speech distortion, especially for the low-level speech spectrum. Simulation results demonstrate improved performance of the proposed technique in terms of the ability to preserve the enhanced speech quality.