AudioMarathon/AudioMarathon

AudioMarathon是一个长上下文音频基准测试数据集，旨在评估多模态LLMs在语音、音乐、环境音频和会议等任务上的性能。该数据集包含11个不同的任务，包括会议摘要、自动语音识别、阅读理解、真实性检测、音乐流派分类、声学场景分类、情感识别、口语命名实体推理、声音事件检测、说话者性别分类和说话者年龄分类。数据集由多个目录组成，如AliMeeting、librispeech-long等，总计包含6204个音频文件和约68.075 GB的音频数据。此外，还提供了一个评审样本子集，包含60个音频文件和2.34 GB的音频数据，用于评审目的。

AudioMarathon is a long-context audio benchmark for evaluating multimodal LLMs on speech, music, environmental audio, and meetings. The dataset includes 11 benchmark tasks spanning meeting summarization, automatic speech recognition, reading comprehension, authenticity detection, music genre classification, acoustic scene classification, emotion recognition, spoken named entity reasoning, sound event detection, speaker gender classification, and speaker age classification. It consists of multiple directories such as AliMeeting, librispeech-long, etc., totaling 6204 audio files and approximately 68.075 GB of audio data. Additionally, a reviewer-facing sample subset is provided, containing 60 audio files and 2.34 GB of audio payload for review purposes.