Generate, Analyze, and Refine: Training-Free Sound Source Localization via MLLM Meta-Reasoning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Park, Subin, Kim, Jung Uk
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911575861886976
author Park, Subin
Kim, Jung Uk
author_facet Park, Subin
Kim, Jung Uk
contents Sound source localization task aims to identify the locations of sound-emitting objects by leveraging correlations between audio and visual modalities. Most existing SSL methods rely on contrastive learning-based feature matching, but lack explicit reasoning and verification, limiting their effectiveness in complex acoustic scenes. Inspired by human meta-cognitive processes, we propose a training-free SSL framework that exploits the intrinsic reasoning capabilities of Multimodal Large Language Models (MLLMs). Our Generation-Analysis-Refinement (GAR) pipeline consists of three stages: Generation produces initial bounding boxes and audio classifications; Analysis quantifies Audio-Visual Consistency via open-set role tagging and anchor voting; and Refinement applies adaptive gating to prevent unnecessary adjustments. Extensive experiments on single-source and multi-source benchmarks demonstrate competitive performance. The source code is available at https://github.com/VisualAIKHU/GAR-SSL.
format Preprint
id arxiv_https___arxiv_org_abs_2604_06824
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Generate, Analyze, and Refine: Training-Free Sound Source Localization via MLLM Meta-Reasoning
Park, Subin
Kim, Jung Uk
Computer Vision and Pattern Recognition
Sound source localization task aims to identify the locations of sound-emitting objects by leveraging correlations between audio and visual modalities. Most existing SSL methods rely on contrastive learning-based feature matching, but lack explicit reasoning and verification, limiting their effectiveness in complex acoustic scenes. Inspired by human meta-cognitive processes, we propose a training-free SSL framework that exploits the intrinsic reasoning capabilities of Multimodal Large Language Models (MLLMs). Our Generation-Analysis-Refinement (GAR) pipeline consists of three stages: Generation produces initial bounding boxes and audio classifications; Analysis quantifies Audio-Visual Consistency via open-set role tagging and anchor voting; and Refinement applies adaptive gating to prevent unnecessary adjustments. Extensive experiments on single-source and multi-source benchmarks demonstrate competitive performance. The source code is available at https://github.com/VisualAIKHU/GAR-SSL.
title Generate, Analyze, and Refine: Training-Free Sound Source Localization via MLLM Meta-Reasoning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.06824