MMSU
收藏资源简介:
MMSU是一个大规模的多任务口语理解和推理基准,由香港中文大学的研究团队创建。该数据集包含5000个经过精心策划的音频-问题-答案三元组,跨越47个不同的任务。数据集系统地融入了广泛的语音现象,包括语音学、韵律、修辞、句法学、语义学和副语言学。MMSU旨在通过评估14个先进的SpeechLLMs来建立口语理解的新标准,并为开发更复杂的人机语音交互系统提供有价值的见解。
MMSU is a large-scale multi-task spoken language understanding and reasoning benchmark developed by the research team at The Chinese University of Hong Kong. This dataset includes 5,000 carefully curated audio-question-answer triples spanning 47 distinct tasks. It systematically incorporates a wide range of speech phenomena, including phonetics, prosody, rhetoric, syntax, semantics, and paralinguistics. MMSU aims to establish new standards for spoken language understanding by evaluating 14 state-of-the-art SpeechLLMs, and provide valuable insights for the development of more sophisticated human-machine speech interaction systems.




