dMelodies
收藏资源简介:
dMelodies是一个专为解耦学习设计的音乐数据集,由佐治亚理工学院音乐技术中心创建。该数据集包含约135万条2小节的单音旋律,每条旋律由九个潜在因素的独特组合生成,涵盖序数、类别和二进制类型。数据集的创建过程遵循简单而多样的独立潜在因素设计原则,确保了数据点的独立性和可区分性。dMelodies数据集主要应用于音乐信息检索和生成音乐模型领域,旨在通过解耦学习提高音乐生成工具的实用性和控制性,解决音乐领域中的解耦表示学习问题。
dMelodies is a music dataset specifically designed for disentangled learning, created by the Georgia Tech Center for Music Technology. This dataset contains approximately 1.35 million 2-bar monophonic melodies, each generated by a unique combination of nine latent factors covering ordinal, categorical, and binary types. The dataset was developed following the principle of simple yet diverse independent latent factor design, which ensures the independence and distinguishability of each data point. The dMelodies dataset is primarily applied in the fields of Music Information Retrieval (MIR) and music generation models, aiming to improve the practicality and controllability of music generation tools via disentangled learning and solve the problem of disentangled representation learning in the music domain.

- 1dMelodies: A Music Dataset for Disentanglement Learning佐治亚理工学院音乐技术中心 · 2020年



