Can Large Audio-Language Models Truly Hear?
Tackling Hallucinations with Multi-Task Assessment and Stepwise Audio Reasoning