Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Human Speech to Ocean Signals: Transferring Speech Large Models for Underwater Acoustic Target Recognition

Mengcheng Huang, Xue Zhou, Chen Xu, Dapeng Man

TL;DR

The paper addresses data scarcity and environmental variability in underwater acoustic target recognition (UATR). It proposes UATR-SLM, which reuses the speech feature pipeline and treats a speech large model as the acoustic encoder with a lightweight classifier, enabling efficient transfer with minimal fine-tuning. On DeepShip and ShipsEar, UATR-SLM achieves in-domain accuracy above 99% and cross-domain accuracy up to 96.67%, demonstrating strong transferability of acoustic priors from speech foundation models to underwater acoustics. This work suggests that speech foundation models encode transferable acoustic priors useful for non-speech domains, offering a data-efficient path for ocean sensing.

Abstract

Underwater acoustic target recognition (UATR) plays a vital role in marine applications but remains challenging due to limited labeled data and the complexity of ocean environments. This paper explores a central question: can speech large models (SLMs), trained on massive human speech corpora, be effectively transferred to underwater acoustics? To investigate this, we propose UATR-SLM, a simple framework that reuses the speech feature pipeline, adapts the SLM as an acoustic encoder, and adds a lightweight classifier.Experiments on the DeepShip and ShipsEar benchmarks show that UATR-SLM achieves over 99% in-domain accuracy, maintains strong robustness across variable signal lengths, and reaches up to 96.67% accuracy in cross-domain evaluation. These results highlight the strong transferability of SLMs to UATR, establishing a promising paradigm for leveraging speech foundation models in underwater acoustics.

From Human Speech to Ocean Signals: Transferring Speech Large Models for Underwater Acoustic Target Recognition

TL;DR

The paper addresses data scarcity and environmental variability in underwater acoustic target recognition (UATR). It proposes UATR-SLM, which reuses the speech feature pipeline and treats a speech large model as the acoustic encoder with a lightweight classifier, enabling efficient transfer with minimal fine-tuning. On DeepShip and ShipsEar, UATR-SLM achieves in-domain accuracy above 99% and cross-domain accuracy up to 96.67%, demonstrating strong transferability of acoustic priors from speech foundation models to underwater acoustics. This work suggests that speech foundation models encode transferable acoustic priors useful for non-speech domains, offering a data-efficient path for ocean sensing.

Abstract

Underwater acoustic target recognition (UATR) plays a vital role in marine applications but remains challenging due to limited labeled data and the complexity of ocean environments. This paper explores a central question: can speech large models (SLMs), trained on massive human speech corpora, be effectively transferred to underwater acoustics? To investigate this, we propose UATR-SLM, a simple framework that reuses the speech feature pipeline, adapts the SLM as an acoustic encoder, and adds a lightweight classifier.Experiments on the DeepShip and ShipsEar benchmarks show that UATR-SLM achieves over 99% in-domain accuracy, maintains strong robustness across variable signal lengths, and reaches up to 96.67% accuracy in cross-domain evaluation. These results highlight the strong transferability of SLMs to UATR, establishing a promising paradigm for leveraging speech foundation models in underwater acoustics.
Paper Structure (16 sections, 4 figures, 5 tables)

This paper contains 16 sections, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Problem Formulation
  4. Progress of Speech Large Models
  5. UATR-SLM Framework
  6. Experiment
  7. Experimental Settings
  8. Datasets and Pre-processing
  9. Model Settings
  10. Evaluation Settings
  11. Experimental Results
  12. Basic Performance
  13. Robustness
  14. Cross-domain generalization
  15. Conclusion and Future Work
  16. ...and 1 more sections

Figures (4)

  • Figure 1: Overview of UATR-SLM. Underwater signals follow the standard speech feature pipeline, are encoded by a speech large model pretrained on massive corpora, and are classified by a lightweight head.
  • Figure 2: Accuracy of different models on variable-length audio clips.
  • Figure 3: T-SNE of ResNet18.
  • Figure 4: T-SNE of UATR-SLM.