Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-driven Joint Detection and Localization of Acoustic Reflectors

H. Nazim Bicer, Cagdas Tuna, Andreas Walther, Emanuël A. P. Habets

TL;DR

The paper addresses the challenge of inferring room geometry when some walls are highly absorptive or distant by jointly detecting and localizing acoustic reflectors. It advances a data-driven approach that extends a prior CRNN reflector localization model to a joint detection/localization framework on-device, using a Radon-domain input and an attention-based loss that weights wall regression by detection estimates. The core contributions are the A-JDL and RA-JDL loss formulations, which enable self-attention to walls that are easier to estimate without requiring explicit detectability labels, and a comprehensive simulated evaluation showing improved localization accuracy for walls that are nearby or highly reflective. This work enhances robust partial room maps and paves the way for scalable, multi-device room geometry inference in complex environments, with practical implications for dereverberation, auralization, and localization tasks.

Abstract

Room geometry inference algorithms rely on the localization of acoustic reflectors to identify boundary surfaces of an enclosure. Rooms with highly absorptive walls or walls at large distances from the measurement setup pose challenges for such algorithms. As it is not always possible to localize all walls, we present a data-driven method to jointly detect and localize acoustic reflectors that correspond to nearby and/or reflective walls. A multi-branch convolutional recurrent neural network is employed for this purpose. The network's input consists of a time-domain acoustic beamforming map, obtained via Radon transform from multi-channel room impulse responses. A modified loss function is proposed that forces the network to pay more attention to walls that can be estimated with a small error. Simulation results show that the proposed method can detect nearby and/or reflective walls and improve the localization performance for the detected walls.

Data-driven Joint Detection and Localization of Acoustic Reflectors

TL;DR

The paper addresses the challenge of inferring room geometry when some walls are highly absorptive or distant by jointly detecting and localizing acoustic reflectors. It advances a data-driven approach that extends a prior CRNN reflector localization model to a joint detection/localization framework on-device, using a Radon-domain input and an attention-based loss that weights wall regression by detection estimates. The core contributions are the A-JDL and RA-JDL loss formulations, which enable self-attention to walls that are easier to estimate without requiring explicit detectability labels, and a comprehensive simulated evaluation showing improved localization accuracy for walls that are nearby or highly reflective. This work enhances robust partial room maps and paves the way for scalable, multi-device room geometry inference in complex environments, with practical implications for dereverberation, auralization, and localization tasks.

Abstract

Room geometry inference algorithms rely on the localization of acoustic reflectors to identify boundary surfaces of an enclosure. Rooms with highly absorptive walls or walls at large distances from the measurement setup pose challenges for such algorithms. As it is not always possible to localize all walls, we present a data-driven method to jointly detect and localize acoustic reflectors that correspond to nearby and/or reflective walls. A multi-branch convolutional recurrent neural network is employed for this purpose. The network's input consists of a time-domain acoustic beamforming map, obtained via Radon transform from multi-channel room impulse responses. A modified loss function is proposed that forces the network to pay more attention to walls that can be estimated with a small error. Simulation results show that the proposed method can detect nearby and/or reflective walls and improve the localization performance for the detected walls.
Paper Structure (11 sections, 8 equations, 3 figures, 1 table)

This paper contains 11 sections, 8 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Proposed Method
  4. Radon transform
  5. Network architecture
  6. Attention-based learning
  7. Performance Evaluation
  8. Data, Training and Metrics
  9. Results
  10. Conclusion
  11. Acknowledgments

Figures (3)

  • Figure 1: CRNN architecture. Detection branch (left) outputs detection scores for each wall, and regression branch (right) estimates corresponding wall normals. The regression branch is equal to the model proposed in tuna2023datadriven.
  • Figure 2: 2D floor map estimation examples.
  • Figure 3: Effect of varying absorption coefficient on detection output for different models.