VoxSafeBench This dataset is uploaded as raw files (JSONL + audio), not parquet. Subset: Safety tier1 Split: No jailbreak (8708 samples) Columns: system prompt, clean audio file name, diverse audio file name, transcript, super category, task type, language, query Split: Singleturn jailbreak (2516 samples) Columns: system prompt, audio file name, transcript, source text, super category, jailbreak type, task type, language, query… See the full description on the dataset page: https://huggingface.co/datasets/nips26/VoxSafeBench.
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy