[Paper Review] Reflection-Aware Sound Source Localization

Inkyu An, Myungbae Son+2 more|arXiv (Cornell University)|Nov 21, 2017

Indoor and Outdoor Localization Technologies24 references3 citations

TL;DR

This paper proposes a reflection-aware 3D sound source localization method that models both direct and reflected sound rays using inverse acoustic ray tracing and Monte Carlo localization, achieving an average localization error of 0.8 m in a 7m×7m×3m room—even with moving or non-line-of-sight sources. The method improves accuracy by 40% over direct-ray-only approaches by incorporating higher-order reflections.

ABSTRACT

We present a novel, reflection-aware method for 3D sound localization in indoor environments. Unlike prior approaches, which are mainly based on continuous sound signals from a stationary source, our formulation is designed to localize the position instantaneously from signals within a single frame. We consider direct sound and indirect sound signals that reach the microphones after reflecting off surfaces such as ceilings or walls. We then generate and trace direct and reflected acoustic paths using inverse acoustic ray tracing and utilize these paths with Monte Carlo localization to estimate a 3D sound source position. We have implemented our method on a robot with a cube-shaped microphone array and tested it against different settings with continuous and intermittent sound signals with a stationary or a mobile source. Across different settings, our approach can localize the sound with an average distance error of 0.8m tested in a room of 7m by 7m area with 3m height, including a mobile and non-line-of-sight sound source. We also reveal that the modeling of indirect rays increases the localization accuracy by 40% compared to only using direct acoustic rays.

Motivation & Objective

• To address the challenge of accurate 3D sound source localization in complex indoor environments with obstacles and non-line-of-sight conditions.
• To overcome the limitations of traditional TDOA-based methods that rely on continuous signals and line-of-sight propagation.
• To improve localization accuracy by modeling both direct and reflected sound paths using acoustic ray tracing.
• To enable real-time localization from a single signal frame without requiring signal accumulation across time or positions.
• To support intermittent, moving, and occluded sound sources in dynamic indoor scenes.

Proposed method

• The method reconstructs the indoor environment using a voxel-based octree to represent room geometry.
• It performs inverse acoustic ray tracing to generate direct and reflected sound paths from microphone array signals using TDOA-based direction estimation.
• Reflected rays are traced through multiple bounces (up to fourth order) off walls and ceilings, simulating specular reflections.
• Monte Carlo localization is applied to estimate the 3D source position by sampling possible source locations along the traced acoustic paths.
• The algorithm uses particle filters to track the source, with likelihoods computed based on signal consistency with predicted ray paths.
• The approach operates on a single frame of audio input, enabling real-time performance without signal accumulation.

Experimental results

Research questions

• RQ1 Can a single-frame, reflection-aware sound source localization method achieve high accuracy in complex indoor environments with obstacles?
• RQ2 How does modeling higher-order reflections affect localization accuracy compared to using only direct rays?
• RQ3 To what extent can the method localize moving or intermittent sound sources when direct paths are blocked?
• RQ4 How does the method perform under non-line-of-sight conditions with occlusions?
• RQ5 What is the optimal maximum order of reflections to balance accuracy and computational cost?

Key findings

• The method achieves an average localization error of 0.8 m in a 7 m × 7 m × 3 m room, even with a moving sound source and occlusion.
• Including reflected rays improves localization accuracy by 40% compared to using only direct rays.
• The average error is 0.7 m on the left side (with obstacle) and 0.3 m on the right side, demonstrating robustness in challenging conditions.
• For intermittent signals (toggled every 5 seconds), the mean error is 0.66 m with a standard deviation of 0.29 m, showing strong performance on non-continuous sources.
• The use of up to fourth-order reflections yields meaningful accuracy improvements, especially on the obstacle-affected side, justifying the choice of four as the maximum reflection order.
• Particle filter determinants remain small (<0.1), indicating stable convergence of the localization estimate.