NSynth and Lakh MIDI combined dataset

该数据集由9.53百万个MIDI音频对组成，这些对是通过结合NSynth和Lakh MIDI数据集生成的，并包含了一个增强版本，提供了额外的音色多样性。为了增强音色的多样性，该数据集还使用了数字效果进行增强，包括均衡器、失真和混响。在规模上，该数据集拥有9.53百万对用于训练，以及53万对用于测试，此外，通过额外的增强处理，数据集的大小翻了一番。这项任务的目的是实现零样本乐器克隆和文本到乐器合成。

This dataset consists of 9.53 million MIDI-audio pairs, which are generated by combining the NSynth and Lakh MIDI Datasets, and features an enhanced variant with expanded timbral diversity. To further augment timbral diversity, the dataset is additionally processed with digital audio effects including equalizers, distortion, and reverb. In terms of scale, the dataset includes 9.53 million pairs for training and 530,000 pairs for testing. Furthermore, the total size of the dataset is doubled via additional augmentation processing. The objective of the corresponding task is to achieve zero-shot instrument cloning and text-to-instrument synthesis.