Learning Magnitude Distribution of Sound Fields via Conditioned Autoencoder

2025-06-20

AI Summary

Modeling the magnitude distribution of acoustic transfer functions (ATFs) becomes challenging under spatially sparse microphone layouts where phase information is unreliable or unavailable. Method: This paper proposes a source-receiver position- and frequency-jointly conditioned autoencoder architecture. It explicitly models the latent variable aggregation module as a deep learning generalization of acoustic field basis expansion, integrating joint spatial-frequency embedding with conditional neural networks to achieve end-to-end ATF magnitude mapping. Contribution/Results: Numerical experiments demonstrate that the method achieves high-accuracy ATF magnitude estimation using only a small number of receivers—significantly outperforming conventional orthogonal basis expansion approaches. It establishes an interpretable, highly robust, data-driven modeling paradigm for low-sampling-rate spatial audio reconstruction, bridging physical acoustics priors with deep learning flexibility.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

A learning-based method for estimating the magnitude distribution of sound fields from spatially sparse measurements is proposed. Estimating the magnitude distribution of acoustic transfer function (ATF) is useful when phase measurements are unreliable or inaccessible and has a wide range of applications related to spatial audio. We propose a neural-network-based method for the ATF magnitude estimation. The key feature of our network architecture is the input and output layers conditioned on source and receiver positions and frequency and the aggregation module of latent variables, which can be interpreted as an autoencoder-based extension of the basis expansion of the sound field. Numerical simulation results indicated that the ATF magnitude is accurately estimated with a small number of receivers by our proposed method.

Problem

• Estimating sound field magnitude from sparse measurements
• Handling unreliable phase data in acoustic transfer functions
• Improving spatial audio applications via neural networks

Innovation

• Conditioned autoencoder for sound field estimation
• Neural network aggregates latent variables
• Sparse measurements enable accurate magnitude estimation