Distributed Acoustic Mixture-of-Experts for Spatial Sound Estimation

Accurate spatial acoustic characterization is crucial for immersive audio applications, such as virtual and augmented reality, which require real-time, high-fidelity rendering of sound fields. Traditional methods, including physics-based simulations and centralized deep learning models, face significant challenges in scalability, data efficiency, and generalization under sparse measurement conditions. This paper introduces DAME, a Distributed Acoustic Mixture-of-Experts framework designed for large-area spatial acoustic modeling. DAME decomposes the global sound field reconstruction task into localized subproblems, each handled by a compact neural network expert trained on region-specific data. A lightweight gating network combines the experts’ predictions based on spatial queries, enabling seamless integration of new measurements without full retraining. In addition, the framework supports incremental expansion, allowing the system to grow by incorporating new experts while preserving previously learned knowledge. Evaluated on a benchmark dataset of higher-order ambisonic room impulse responses, DAME significantly outperforms state-of-the-art baselines in terms of reconstruction accuracy, directional estimation, and energy decay preservation. Computational analysis demonstrates its suitability for real-time operation, and a perceptual evaluation confirms superior audio quality and spatial fidelity. The framework, therefore, provides a scalable, data-efficient solution for robust spatial audio reconstruction in challenging sparse-data regimes.