MusicNet 音乐数据集

MusicNet 是一个大型音乐数据集，用以监督和评估音乐研究的机器学习方法。该数据集由 330 张版权免费的古典音乐唱片以及超 100 万个注释标签组成。注明了每张唱片每个音符出现的准确时间、演奏该音符的乐器以及该音符在作品韵律结构中的位置。这些标签经由音乐家评估验证，得出标签的错误率仅为 4% 。

MusicNet is a large-scale music dataset designed to supervise and evaluate machine learning methods for music research. It comprises 330 copyright-free classical music recordings and over 1 million annotated labels, which record the exact occurrence time of each note on every recording, the instrument that played the note, and the position of the note within the rhythmic structure of the corresponding musical work. These labels were validated and assessed by musicians, resulting in a label error rate of only 4%.

搜集汇总

背景与挑战

背景概述

MusicNet是一个专注于古典音乐的大型数据集，包含330张版权免费唱片和超过100万个高精度注释标签，每个标签注明了音符的时间、乐器和韵律位置，错误率仅为4%，适用于音乐相关的机器学习研究和音频处理任务。