SpotSound: Enhancing Large Audio-Language Models with Fine-Grained Temporal Grounding
2026-04-14 • Sound
SoundMultimedia
AI summaryⓘ
The authors created SpotSound, an audio-language model that can better find exactly when a specific event happens in long audio recordings. They improved training by teaching the model not to guess event timings if the event isn't actually there. They also made a tough test, SpotSound-Bench, where events are very short and hidden in background noise, like finding a needle in a haystack. Their model outperforms others on timing tasks while still working well for general audio understanding.
Audio-Language ModelsTemporal GroundingLong-form AudioEvent LocalizationTraining ObjectivesBenchmark DatasetBackground NoiseClip-level SupervisionAudio Event Detection
Authors
Luoyi Sun, Xiao Zhou, Zeqian Li, Ya Zhang, Yanfeng Wang, Weidi Xie
Abstract
Large Audio-Language Models (ALMs) have recently demonstrated remarkable capabilities in holistic audio understanding, yet they remain unreliable for temporal grounding, i.e., the task of pinpointing exactly when an event occurs within long-form audio. This limitation stems from two factors: training data dominated by clip-level supervision lacking precise timestamps, and benchmarks that fail to simulate real-world scenarios where short events are obscured by dense background sounds. In this paper, we introduce SpotSound, an audio language model designed for grounding audio events. SpotSound incorporates a novel training objective, specifically designed to suppress hallucinated timestamps for events absent from the input. Additionally, we present SpotSound-Bench, a challenging temporal grounding benchmark where target events occupy less than ~10\% of each clip, creating a rigorous `needle-in-a-haystack' evaluation. Experiments demonstrate that SpotSound achieves state-of-the-art results on temporal grounding benchmarks while maintaining robust performance across general downstream audio-language tasks. Code, models and benchmark are released on https://loiesun.github.io/spotsound/