Multiple far noise suppression in a real environment using transfer-function-gain NMF

Conventional noise suppression methods based on array signal processing use phase information and control the directivity of noises. However, such methods can hardly suppress so-called background noise, whose arrival direction cannot be specified. Thus, multiple far noise suppression based on transfer-function-gain non-negative matrix factorization (NMF) has been proposed as a method that can suppress such background noise. Its effectiveness has been confirmed by an experimental simulation using convolutional mixtures; however, it has not been verified that it is practical in a real environment. Thus, in this paper, we examine the performance of this method by recording a target and multiple far noises with asynchronous microphones in a real environment. We confirm that this method can suppress far noises in a real environment with diverse distances between microphones and interference sources.