Slakh
收藏资源简介:
Slakh数据集是由三菱电机研究实验室和西北大学的交互音频实验室合作创建的,专注于音乐源分离研究。该数据集包含2100首歌曲,总计145小时的混合音频,由专业级的基于样本的虚拟乐器从Lakh MIDI数据集(LMD)生成。Slakh数据集通过精细控制合成参数,提供了比现有数据集更多的数据量,有助于提升音乐信号分析的性能,特别是对于音乐源分离技术的发展。此外,Slakh还支持多种音乐信号分析任务,如乐器识别和音乐转录,为解决音乐源分离中的数据稀缺问题提供了新的解决方案。
Slakh Dataset was co-developed by Mitsubishi Electric Research Laboratories and the Interactive Audio Lab of Northwestern University, with a focus on music source separation research. This dataset contains 2100 songs totaling 145 hours of mixed audio, which was generated using professional sample-based virtual instruments sourced from the Lakh MIDI Dataset (LMD). By finely controlling synthesis parameters, the Slakh Dataset offers a larger volume of data than existing datasets, helping to enhance the performance of musical signal analysis, especially for the advancement of music source separation technologies. Furthermore, Slakh supports a variety of musical signal analysis tasks such as instrument recognition and music transcription, providing a novel solution to the data scarcity problem in music source separation.

- Slakh数据集首次发表,由Jesse Engel等人提出,旨在为音乐生成和音频分离任务提供一个大规模、多样的数据集。
- Slakh数据集在多个国际会议和期刊上被广泛引用,成为音乐信息检索领域的重要基准数据集。
- Slakh数据集的应用扩展到深度学习模型的训练,特别是在自动音乐生成和音频分离任务中取得了显著成果。
- Slakh数据集的版本更新,增加了更多的音频样本和多样性,进一步提升了其在音乐信息检索和生成任务中的应用价值。



