GiantMIDI-Piano: A Large-Scale MIDI Dataset for Classical Piano Music

Symbolic music datasets are important for information retrieval and musical analysis. However, there is a lack of large-scale symbolic classical piano music. In this article, we describe the creation GiantMIDI-Piano (GP) dataset containing 38,700,838 transcribed notes 10,855 unique solo works composed by 2,786 composers. We extract names composers from International Music Score Library Project (IMSLP). search download their corresponding audio recordings Internet. further create curated subset 7,236 1,787 where titles downloaded contain surnames apply convolutional neural network to detect works. Then, transcribe those into Musical Instrument Digital Interface (MIDI) files using high-resolution transcription system. Each MIDI file contains onset, offset, pitch, velocity attributes pedals. includes 90% live performance 10% sequence input files. analyse statistics show pitch class, interval, trichord, tetrachord frequencies six different eras that can be used evaluate quality in terms detection F1 scores, metadata accuracy, error rates. release source code acquiring at https://github.com/bytedance/GiantMIDI-Piano.