AIR-Bench

该数据集名为AIR-Bench，是首个被广泛采用的评估基准，旨在评估语言音频模型（LALMs）在处理包括人类语音、自然声音和音乐在内的各种音频信号时的理解能力。其任务是对LALMs在音频信号范围内的理解能力进行评估。

This dataset, named AIR-Bench, is the first widely adopted evaluation benchmark designed to assess the comprehension capabilities of Language-Audio Models (LALMs) when processing a diverse array of audio signals including human speech, natural sounds, and music. The core task of this benchmark is to evaluate the comprehension performance of LALMs across the full scope of audio signals.