Development and evaluation of spherical microphone array-enabled systems for immersive multi-user environments

Abstract

As viable applications for human-scale immersive spaces continue to be discovered, the need for reactive elements to enable effective sensing becomes increasingly prevalent. In this thesis, a spherical microphone array, or higher-order Ambisonic (HOA) array, is examined in the context of immersive, multi-user environments, to assess feasibility of real-time operation for sound-source tracking, isolation, and telematic broadcast. The design and operation of a prototype HOA array is reviewed, including static and adaptive beamforming methods for sound source isolation. Several contemporary techniques for direction of arrival (DOA) estimation are explored, with their benefits and drawbacks examined. A method for DOA estimation which addresses the shortcomings of existing techiques, the Sparse Iterative Search (SIS), is designed and evaluated against existing techniques, using a computational model and laboratory trials. This method is extended using a clustering method to demonstrate acoustic source identification. To avoid limitations with acoustic-only source identification techniques, a multiple-sensor paradigm is designed to identify multiple users, associate voice activity to individual users, and isolate voice activity using the fusion of streaming data from an HOA array and array of depth-sensing cameras with low latency. This sensor modality is tested using a computational model and tests in a lab environment. This document records the theory, process, and execution of this research study.