This dataset contains transcriptions of the electric guitar performance of 240 tablatures, rendered with different tones. The goal is to contribute to automatic music transcription (AMT) of guitar music, a technically challenging task.

Activity signals were captured by attaching a special hexaphonic pickup to each string of an electric guitar and using a JUCE program to control a digital audio workstation (DAW) to automatically re-render the audio recordings of the “Direct Input” (DI) using different amplifiers (Amps), including low-gain amps and high-gain ones. A new collecting pipeline was employed to reduce the effort of manual inspection. The final dataset contains six copies of 118 minutes of guitar playing, each copy being associated with a different timbre. The new dataset, named “EGDB,” is constructed in this way to account for the diverse timbre associated with electric guitar. Having multiple guitar tones makes it possible to test a trained model on held-out unseen tones for generalizability.

( Image Source: Frame Harirak )