2020/11/03 by Nicolas Furnon, Furnon, Nicolas, Serizel, Romain +4 · 5 citations
Computer Science · Engineering · Neuroscience · #Advanced Adaptive Filtering Techniques #FOS: Electrical engineering #Hearing Loss and Rehabilitation #Signal Processing (eess.SP) #Speech and Audio Processing #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2011.01714
openalex publication_date 2020/11/03 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Deep neural network (DNN)-based speech enhancement algorithms in microphone\narrays have now proven to be efficient solutions to speech understanding and\nspeech recognition in noisy environments. However, in the context of ad-hoc\nmicrophone arrays, many challenges remain and raise the need for distributed\nprocessing. In this paper, we propose to extend a previously introduced\ndistributed DNN-based time-frequency mask estimation scheme that can\nefficiently use spatial information in form of so-called compressed signals\nwhich are pre-filtered target estimations. We study the performance of this\nalgorithm under realistic acoustic conditions and investigate practical aspects\nof its optimal application. We show that the nodes in the microphone array\ncooperate by taking profit of their spatial coverage in the room. We also\npropose to use the compressed signals not only to convey the target estimation\nbut also the noise estimation in order to exploit the acoustic diversity\nrecorded throughout the microphone array.\n