Qwen-2-Audio

The 'Qwen-2-Audio' dataset is used to evaluate hallucination detection methods in Speech Large Language Models (SpeechLLMs) by analyzing attention-derived metrics during inference.

Papers using Qwen-2-Audio (3)

Continuous Audio Thinking for Large Audio Language Models2026Detecting Hallucinations in SpeechLLMs at Inference Time Using Attention Maps2026Extending Audio Context for Long-Form Understanding in Large Audio-Language Models2025