Lakh-Ilya dataset

该数据集是一个大规模的MIDI与音频对齐的木吉他数据集，它通过将Lakh MIDI数据集中的MIDI数据与从木吉他库合成的音频进行对齐而创建。其中，90%的曲目用于训练，10%用于验证。音频被下采样至16kHz，并与梅尔频谱图进行逐帧对齐。该数据集包含超过100小时的数据，旨在为木吉他合成预训练模型的任务提供支持。

This dataset is a large-scale aligned acoustic guitar dataset that pairs MIDI data with audio. It was constructed by aligning the MIDI tracks from the Lakh MIDI Dataset with audio synthesized from an acoustic guitar library. Specifically, 90% of the tracks are allocated for model training, while the remaining 10% are reserved for validation. The audio is downsampled to 16 kHz and aligned frame-by-frame with mel-spectrograms. This dataset contains over 100 hours of data, and is designed to support pre-trained model development tasks for acoustic guitar synthesis.