Speech Denoising Using Bayesian NMF with Online Base Update

Abstract

A new online non-negative matrix factorization (NMF) based speech denoising method was proposed in this paper. To achieve an efficient model for the temporal dependencies of speech and noise, and improve the robustness for the actual nonstationary noisy environments, the Bayesian NMF was extended to the proposed model and a new noise basis matrix online update method was exploited. Firstly, the speech basis matrix was pretrained off-line using the Bayesian NMF method. In speech denoising stage, the noise basis matrix was continuously updated utilizing the noise frames in the noisy observation with the Bayesian NMF. The noise basis matrix was initialized via a pretrained universal noise Bayesian NMF model and the noise data for the matrix adaption were selected using a likelihood ratio test (LRT) speech decision criterion. Then the updated noise basis matrix and the pre-trained speech basis matrix were employed to the enhancement of the noisy signal. The experiment results show that the proposed method outperforms the comparison denoising algorithms in terms of objective measurement.