Unsupervised noise removal technique based on constrained NMF
V. Anil Kumar, Chevula Venkata Rama Rao · IET Signal Processing · 2017
This paper presents a novel approach to mitigate background interference noise in a noisy audio environment based on unsupervised speech enhancement without the prior knowledge of speech as well as noise signal. In this scheme, a noisy speech signal is represented by a non‐negative matrix factorisation (NMF) with sparseness constraints. Using NMF with constraints, noise density spectrum is estimated to design the spectral subtraction method. In the earlier method of NMF, correlation between the noise and the audio signals was not considered. The proposed method explores the correlation between the noise and speech signals. A sparse constraint NMF‐based spectral subtraction method is designed to eliminate the noise. The performance is evaluated by considering various noise environments in terms of quality and intelligibility measures. Simulation results highlight the improvement in the enhanced speech with the suggested approach.