GM Dataset

The GM Dataset is a collection of guitar performance recordings paired with detailed MIDI transcriptions, designed to advance research in music information retrieval, automatic transcription and related fields.The dataset includes a wide range of songs from different genres, including classical, reggae, jazz and pop. It also captures different guitar techniques, such as arpeggios, solos, chord progressions and different playing styles (e.g. with or without a pick). The MIDI transcriptions are organised on a string-by-string basis, providing string-specific data essential for tablature generation and transcription accuracy.Dataset Composition:Rock: 4 recordings (7:12) of riffs, solos and chord progressions.Jazz/Blues: 5 takes (6:40) including riffs, solos, comping and arpeggios.Pop: 5 tracks (6:58) of chord progressions and arpeggios.Reggae: 2 samples (1:57) of a riff and a chord progression.Classical: 2 takes (2:42) with an arpeggio and a solo.This dataset is a valuable resource for overcoming the challenges of guitar transcription and paves the way for innovations in audio processing and machine learning applications.