Hearing-aids system using distributed assistive device and blind speech extraction method under diffuse noise

To perform a smooth dialogue using the hearing-aids system in various acoustical environments, it is necessary to develop a real-time speech extraction method for hearing aids under diffuse noise conditions even with a limited computational resource. In recent years, almost everyone has a smartphone; thus, we utilize this as an assistive microphone array. For these reasons, we aim at extracting the target speech under diffuse noise conditions with distributed microphone arrays and a limited computational resource. We previously proposed a real-time speech extraction framework based on spatially regularized independent low-rank matrix analysis and rank-constrained spatial covariance matrix estimation. In this paper, we apply our method to hearing-aids systems with an assistive device and evaluate the performance. In experiments, we used eight-channel distributed microphone arrays, which consist of three microphones equipped with each ear and two microphones equipped with an assistive device simulating a smartphone. First, we confirmed that the real-time method using some spatial prior information improves the performance compared to that performed blindly with a high computational resource. Second, considering the constraints of a limited computational resource, we investigated the number and arrangement of microphones to achieve superior performance and confirmed the effectiveness of the assistive device.