Abstract:

This paper focuses on using the single-channel source separation techniques to improve the quality of the desired speech in the real-world environment where the speech signal is corrupted by high-level background noise, and especially when there is no source-specific training data. We propose a solution combining the Nonnegative Matrix Factorization model (NMF) with mixed group sparsity constraints to separate the speech signal from the single - channel audio signal with high ambient noise. Experiment result over mixtures containing different real-world noises confirms the effectiveness of the proposed algorithm.