📖 Biography

I am a 3rd-year PhD student at the Department of Data Science & AI, Monash University (2023 - now), supervised by Prof. Gholamreza Haffari, Dr. Ehsan Shareghi and Dr. Lizhen Qu.

My research interest focuses on the safety issues on audio LMMs and speech-specific risks. We red team current audio LMMs to reveal potential vulnerabilities and explore effective safeguarding mechanism for constructing safer audio LMMs.

🔥 News

• 2025.08:  🎉🎉 Our paper “Reshaping Representation Space to Balance the Safety and Over-rejection in Large Audio Language Models” has been accepted to EMNLP 2025. The code and dataset are coming soon at here.
• 2025.07:  🎉🎉 Our paper “Jigsaw Puzzles: Splitting Harmful Questions to Jailbreak Large Language Models in Multi-turn Interactions” has been accepted to COLM 2025.
• 2025.01:  📚📚 The repository for “Audio Is the Achilles’ Heel: Red Teaming Audio Large Multimodal Models” is available at here.
• 2025.01:  🎉🎉 Our paper “Audio Is the Achilles’ Heel: Red Teaming Audio Large Multimodal Models” has been accepted to NAACL 2025.
• 2024.12:  📚📚 Our codes for “Jigsaw Puzzles: Splitting Harmful Questions to Jailbreak Large Language Models in Multi-turn Interactions” are available at here.
• 2024.11:  📚📚 Our speech-specific risk dataset is available at here.
• 2024.09:  🎉🎉 Our paper “Towards Probing Speech-Specific Risks in Large Multimodal Models: A Taxonomy, Benchmark, and Insights” has been accepted to EMNLP 2024.

📝 Publications

• [EMNLP 2025] Reshaping Representation Space to Balance the Safety and Over-rejection in Large Audio Language Models

  Hao Yang, Lizhen Qu, Ehsan Shareghi, Gholamreza Haffari

• [NAACL 2025] Audio Is the Achilles’ Heel: Red Teaming Audio Large Multimodal Models

  Hao Yang, Lizhen Qu, Ehsan Shareghi, Gholamreza Haffari

• [COLM 2025] Jigsaw Puzzles: Splitting Harmful Questions to Jailbreak Large Language Models in Multi-turn Interactions

  Hao Yang, Lizhen Qu, Ehsan Shareghi, Gholamreza Haffari

• [INTERSPEECH 2025] Continual Speech Learning with Fused Speech Features

  Guitao Wang, Jinming Zhao, Hao Yang, Guilin Qi, Tongtong Wu, Gholamreza Haffari

• [EMNLP 2024] Towards Probing Speech-Specific Risks in Large Multimodal Models: A Taxonomy, Benchmark, and Insights

  Hao Yang, Lizhen Qu, Ehsan Shareghi, Gholamreza Haffari

• [INTERSPEECH 2023] Investigating pre-trained audio encoders in the low-resource condition

  Hao Yang, Jinming Zhao, Gholamreza Haffari, Ehsan Shareghi

• [EMNLP 2022-Findings] Self-supervised rewiring of pre-trained speech encoders: Towards faster fine-tuning with less labels in speech processing

  Hao Yang, Jinming Zhao, Gholamreza Haffari, Ehsan Shareghi

• [EMNLP 2022-Findings] RedApt: An Adaptor for wav2vec 2 Encoding: Faster and Smaller Speech Translation without Quality Compromise

  Jinming Zhao, Hao Yang, Gholamreza Haffari, Ehsan Shareghi

• [INTERSPEECH 2022] M-adapter: Modality adaptation for end-to-end speech-to-text translation

  Jinming Zhao, Hao Yang, Ehsan Shareghi, Gholamreza Haffari