DiffusionRIR: Room Impulse Response Interpolation using Diffusion Models

2025-04-29

AI Summary

High-resolution room impulse response (RIR) measurement is costly and impractical for large-scale or densely sampled scenarios. To address this, this work introduces denoising diffusion probabilistic models (DDPMs) to RIR spatial interpolation—the first such application. Methodologically, RIRs are structured as two-dimensional tensors, enabling image-like modeling and generative reconstruction via diffusion processes. Unlike conventional cubic spline interpolation—which suffers from linearity and locality constraints—our approach captures global, non-linear spatial correlations in RIR fields. Experimental results demonstrate substantial improvements in reconstruction fidelity for sparsely spaced microphone arrays: normalized mean square error (NMSE) decreases by over 40%, and cosine similarity increases by more than 35%. The method achieves high-accuracy RIR interpolation across spatial intervals ranging from centimeter- to meter-scale, enabling practical, cost-effective acquisition of high-resolution RIRs without dense physical sampling.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Matrix & Tensor Methods
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Room Impulse Responses (RIRs) characterize acoustic environments and are crucial in multiple audio signal processing tasks. High-quality RIR estimates drive applications such as virtual microphones, sound source localization, augmented reality, and data augmentation. However, obtaining RIR measurements with high spatial resolution is resource-intensive, making it impractical for large spaces or when dense sampling is required. This research addresses the challenge of estimating RIRs at unmeasured locations within a room using Denoising Diffusion Probabilistic Models (DDPM). Our method leverages the analogy between RIR matrices and image inpainting, transforming RIR data into a format suitable for diffusion-based reconstruction. Using simulated RIR data based on the image method, we demonstrate our approach's effectiveness on microphone arrays of different curvatures, from linear to semi-circular. Our method successfully reconstructs missing RIRs, even in large gaps between microphones. Under these conditions, it achieves accurate reconstruction, significantly outperforming baseline Spline Cubic Interpolation in terms of Normalized Mean Square Error and Cosine Distance between actual and interpolated RIRs. This research highlights the potential of using generative models for effective RIR interpolation, paving the way for generating additional data from limited real-world measurements.

Problem

• Estimating RIRs at unmeasured locations using diffusion models
• Interpolating RIRs for high spatial resolution with limited measurements
• Improving RIR reconstruction accuracy compared to traditional interpolation methods

Innovation

• Uses Denoising Diffusion Probabilistic Models (DDPM)
• Transforms RIR data into image-like format
• Outperforms Spline Cubic Interpolation significantly