musical_distribution_shift
收藏资源简介:
musical_distribution_shift数据集由约翰内斯·开普勒大学林茨计算感知研究所与LIT AI实验室创建,旨在评估自动音乐转录系统在不同音乐分布偏移下的性能。数据集包括MIDI文件和真实钢琴录音,分为不同音乐流派和随机组合的子集,以测试系统的泛化能力。数据集的创建过程包括MIDI文件的收集、合成及在统一声学条件下的录制。该数据集主要用于研究自动音乐转录系统中的分布偏移问题,特别是音乐流派和声音特性对系统性能的影响。
The Musical_Distribution_Shift dataset was developed by the Institute of Computational Perception, Johannes Kepler University Linz and the LIT AI Lab, with the goal of evaluating the performance of automatic music transcription systems under various musical distribution shifts. This dataset comprises MIDI files and real piano recordings, and is split into subsets based on different music genres and random combinations to test the generalization capabilities of such systems. The dataset creation process includes the collection, synthesis of MIDI files, and recording under uniform acoustic conditions. This dataset is primarily utilized to research distribution shift issues in automatic music transcription systems, especially the influence of music genres and acoustic characteristics on system performance.
音乐分布偏移
数据集简介
该数据集用于IWSSPA 2024研讨会提交的论文《量化自动音乐转录系统中的语料库偏差问题》,研究自动音乐转录系统中的音乐分布偏移现象。
数据状态
数据集正在整理中。




