Speech Enhancement Using Semi-Supervised Learning

Speech enhancement using machine learning algorithm is one of the research problems in signal processing. The goal of the research is to enhance the speech signals, their by improving quality and intelligibility voice signals that are corrupted by real world environmental noise. In this paper we consider semi-supervised machine learning algorithm to improve the quality of speech signal corrupted by environmental disturbance. Most of the environmental disturbances are non-stationary i.e. the effect of noise is not uniform for all spectral components. In the proposed algorithm the system training is done using a set of speech and noise data base. Parameters are derived by calculating system performance. These parameters are used to enhance the speech signal. The results that are obtained show a considerable improvement in SNR by 5% to 8% as compared to that of conventional methods.