基于SHS-100K增强的音乐相似识别特征数据集

基于SHS-100K增强的音乐相似识别特征数据集主要面向音乐信息检索中翻唱识别研究以及数字音乐产品版权保护产业发展需求建设，基于开源的SHS-100K翻唱数据集进行数据抽取，音乐下载，侵权类型增强处理，使用预训练模型crema主要生成了原有下载音乐以及增强音乐能够体现基调不变性的cremaPCP特征。数据集共计超过6万个音乐特征文件，占据约5GB大小的存储空间。The SHS-100K-enhanced music similarity recognition feature dataset is developed to address the research needs of cover song recognition in music information retrieval (MIR) and the industrial development requirements of digital music product copyright protection. It is constructed based on the open-source SHS-100K cover song dataset through data extraction, music downloading, and infringement-type data augmentation processing. Using the pre-trained model CreMA, we primarily generate the cremaPCP features that retain the key invariance of both the originally downloaded music and the augmented music. The dataset contains more than 60,000 music feature files and occupies approximately 5 GB of storage space.

搜集汇总

背景与挑战

背景概述

该数据集是基于SHS-100K增强的音乐相似识别特征数据集，主要用于音乐信息检索中的翻唱识别研究和数字音乐产品版权保护。数据集通过预训练模型crema生成了超过6万个音乐特征文件，总大小约为5GB，重点提取了体现基调不变性的cremaPCP特征。