HumTrans

HumTrans数据集是由腾讯PCG的ARC Lab和Foundational Technology Center创建，专为哼唱旋律转录设计。该数据集包含500首不同风格和语言的音乐作品，分为1000个音乐片段，总音频时长约56.22小时，是目前最大的公开哼唱数据集。数据收集过程中，10名音乐专业或乐器演奏熟练的大学生通过专用网站界面录制每个片段两次。数据集不仅适用于旋律转录，还可用于基于哼唱的音乐生成等下游任务。The HumTrans dataset was created by ARC Lab and Foundational Technology Center of Tencent PCG, and is specifically designed for humming melody transcription. This dataset contains 500 musical works of diverse styles and languages, which are split into 1000 audio segments with a total duration of approximately 56.22 hours, making it the largest publicly available humming dataset to date. During the data collection process, 10 college students who are either music majors or proficient in musical instruments recorded each segment twice via a dedicated website interface. In addition to melody transcription, the dataset can also be applied to downstream tasks such as humming-based music generation.

搜集汇总

背景与挑战

背景概述

HumTrans数据集是由腾讯PCG的ARC Lab和Foundational Technology Center创建，专为哼唱旋律转录设计的公开数据集，包含500首音乐作品、1000个片段，总时长约56.22小时，是目前最大的哼唱数据集。数据由10名音乐专业学生通过专用界面录制，适用于旋律转录和基于哼唱的音乐生成等下游任务。