Enhancing 3-D Sound Event Localization and Detection With Distance Estimation Using Reverberation and Spatial Coherence Features

摘要

Sound Event Localization, Detection, and Distance Estimation (3D SELD) is pivotal for applications such as acoustic monitoring, surveillance, and robotic navigation. Despite significant advances in Sound Event Localization and Detection (SELD), achieving accurate Sound Distance Estimation (SDE) remains challenging due to complex real-world reverberation and the limited efficacy of existing methods. To address these challenges, we propose the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Coherence and Direct-Path Dominance (CDPD)</i> feature, specifically designed to overcome current limitations in SDE accuracy. By explicitly integrating complementary spatial coherence and reverberation cues into conventional SELD features, the CDPD feature captures robust distance-related information even under challenging reverberant conditions. Experimental results on the Sony-TAu Realistic Spatial Soundscapes 2023 (STARSS23) dataset demonstrate that incorporating the CDPD feature significantly enhances overall 3D SELD performance, reducing the class-dependent relative distance error by up to 6.07% and improving the location-dependent F-score by up to 8.90%. Comparative analyses on the DCASE Challenge 2024 Task 3 validation set further confirm that our approach attains competitive performance while requiring fewer training resources. These findings highlight the potential of leveraging coherence and direct-path dominance cues to advance 3D SELD in real-world reverberant environments.