Summary

This repository, "Awesome Trustworthy Audio-LLMs," serves as a curated reading list and resource hub for researchers and practitioners focused on the trustworthiness of Audio Large Language Models (Audio-LLMs). It addresses critical aspects such as safety, robustness, fairness, privacy, interpretability, security, and hallucination, aiming to foster a future where Audio-LLMs are reliable and trustworthy. The collection provides a structured overview of the rapidly evolving field.

How It Works

The project functions as a comprehensive, community-driven collection of research papers, benchmarks, datasets, and open resources. It meticulously categorizes these materials under key trustworthiness domains, including Safety (with sub-categories like Jailbreak, Alignment, Deepfake, Prompt Injection, Defense), Security (Adversarial Examples, Attack, Poison & Backdoor), Privacy, Interpretability, Fairness, Hallucination, and Robustness. This organized structure facilitates efficient navigation and discovery of relevant work in trustworthy audio intelligence.

Quick Start & Requirements

This repository is a curated collection of research resources and does not provide direct installation or execution commands for a software project. It serves as a reference guide rather than a runnable tool.

Highlighted Details

• Covers a broad spectrum of trustworthiness concerns for Audio-LLMs, including safety, security, privacy, interpretability, fairness, hallucination, and robustness.
• Features specific research collections and highlights notable papers such as "ChronosAudio: A Comprehensive Long-Audio Benchmark," "AHa-Bench: Benchmarking Audio Hallucinations in Large Audio-Language Models," and "Hidden in the Noise: Unveiling Backdoors in Audio LLMs."
• The collection is continuously updated with recent papers, with updates logged for various periods in 2025 and 2026.

Maintenance & Community

Contributions are welcomed from researchers and practitioners via email to kaiwenluo74@gmail.com or by submitting paper links in the issues section. Discussion channels include a WeChat Group and a dedicated Discord Group for TALLM. The project is organized by Kevin Luo, Zhenhong Zhou, Liang Lin, Yuting Ruan, Yuanhe Zhang, and Tianyu Shao.

Licensing & Compatibility

The provided README does not specify a software license. This omission requires clarification for any potential downstream use or integration.

Limitations & Caveats

As a curated reading list, this repository is a continuously evolving collection rather than a deployed software system. The absence of a specified license is a significant caveat for adoption and commercial use. The project appears to be actively maintained with regular updates to its resource collection.