Distributed Multiarray Noise Reduction With Online Estimation Of Masks And Spatial Filters

A rapid increase in popularity of smart devices equipped with acoustic sensors enables the design of speech enhancement methods for distributed setups. In this work, we present a distributed noise reduction scheme in which the time-frequency masks and spatial filters are estimated at the nodes based on the locally available microphone signals and the compressed signals received from other nodes, where each node transmits only a single signal. The proposed block processing facilitates online estimation of masks and distributed spatial filters in an interchanged fashion.
The results of performed numerical experiments indicate that the proposed online distributed noise reduction scheme performs similarly to the centralized approach, in which signals of all microphones of distributed arrays are available for joint processing, and it significantly outperforms the local approach in which only the local microphone signals are available for the estimation of masks and spatial filters.