Speech enhancement using non negative matrix factorization and enhanced NMF

Speech enhancement is a dominant research area, which is used to augment the degraded speech. This paper is based on speech enhancement using Bayesian Formulation of Nonnegative Matrix Factorization algorithm with Hidden Markov Model (BNMF HMM) for supervised speech enhancement. This paper, explore a new class of unsupervised speech de-noising algorithm known as Enhanced NMF (ENMF), which is used to develop a speech enhancement system neither the speaker identity nor the type of noise is known in advance. NMF methods are used for source separation. Extensive experiments are carried out to examine the performance of the proposed method under different circumstances. Furthermore, this work contrasts the recital of the developed algorithms with state of the art speech enhancement schemes using various objective and subjective measures. The experimental results show that Enhanced NMF (ENMF) method is more efficient than other conservative methods by using Signal to Noise Ratio (SNR) in dB, Signal to Distortion Ratio (SDR) in dB and Perceptual Evaluation of Speech Quality (PESQ) in Mean Opinion Score.