DISCO-10M
收藏资源简介:
DISCO-10M是由苏黎世联邦理工学院创建的大型音乐数据集,包含15296232条音乐数据,远超现有音乐数据集的规模。该数据集通过多阶段过滤过程确保数据质量,包括基于文本描述和音频嵌入的相似性。此外,数据集还提供了预计算的CLAP嵌入,便于直接应用于各种下游任务。DISCO-10M的目标是民主化和促进新研究,帮助推动音乐领域机器学习模型的创新发展。数据集内容丰富,覆盖多种音乐类型和来源,创建过程中采用了严格的筛选和匹配机制。应用领域广泛,旨在解决音乐分析、推荐系统和音乐创作中的问题。
DISCO-10M is a large-scale music dataset developed by ETH Zurich, containing 15,296,232 music samples, which far exceeds the scale of existing music datasets. This dataset ensures data quality through a multi-stage filtering process, including similarity checks based on textual descriptions and audio embeddings. Additionally, the dataset provides pre-computed CLAP embeddings to enable direct application across various downstream tasks. The core objective of DISCO-10M is to democratize access to and advance music-related research, facilitating the innovative development of machine learning models in the music domain. The dataset boasts rich content covering diverse music genres and sources, with strict screening and matching mechanisms adopted during its construction. It has a wide range of application scenarios, aiming to address key issues in music analysis, recommendation systems, and music creation.

- 1DISCO-10M: A Large-Scale Music Dataset苏黎世联邦理工学院 · 2023年



