Evaluation of Multichannel Hearing Aid System by Rank-Constrained Spatial Covariance Matrix Estimation

In a noisy environment, speech extraction techniques make hearing aid systems more effective and practical. Blind source separation (BSS) is suitable for hearing aids because it can be employed without any a priori spatial information. Among many BSS methods, independent low-rank matrix analysis (ILRMA) achieves high-quality separation performance. In a diffuse-noise environment, however, ILRMA cannot suppress the noise since it is based on the determined situation. On the other hand, rank-constrained spacial covariance matrix (SCM) estimation overcomes the problem. This method utilizes spatial parameters accurately estimated by ILRMA and compensates for the deficiency of the spatial basis of diffuse noise. The application of BSS methods to a multichannel binaural hearing aid system with a smartphone has never been studied in detail thus far. To clarify the efficacy of the BSS methods in real environments, we record real sounds by constructing a hearing aid system with a dummy head and a smartphone. In this study, we investigate the applicability of BSS for a multichannel binaural hearing aid system with microphones on a smartphone. Furthermore, we apply ILRMA and the rank-constrained SCM estimation to the recorded data and evaluate these methods in terms of their separation performance.