Recognition of physically structured underwater signals by general-purpose audio-language models

Determine whether existing general-purpose large audio-language models can reliably recognize physically structured underwater acoustic signals without task-specific adaptation.

Background

The paper notes that existing large audio-LLMs are primarily designed for speech, music, and environmental sounds rather than underwater acoustics, whose vessel-radiated signals contain physically structured tonal, modulation, and machinery-related patterns. Their ability to transfer to fine-grained underwater recognition is therefore unresolved. The cited UA-Bench results show low leaf-level accuracy for nine unadapted models, motivating the development of physics-aware adaptation, but they do not establish whether general-purpose models can achieve reliable recognition in this domain.

References

In addition, existing general-purpose LALMs are primarily designed to understand speech, music, and environmental sounds, and their ability to recognize physically structured underwater signals remains uncertain.

— From Signals to Evidence: Interpretable Underwater Acoustic Target Sensing with Large Language Models - Part I: Framework and Recognition  (2610.22064 - Zhang et al., 15 Oct 2026) in Section 2, subsection “Large Audio-Language Models and Cross-Modal Alignment”