Guitar-TECHS: An Electric Guitar Dataset Covering Techniques, Musical Excerpts, Chords and Scales Using a Diverse Array of Hardware

Please cite this paper if you use Guitar-TECHS:

Pedroza, Hegel, et al. "Guitar-TECHS: An Electric Guitar Dataset Covering Techniques, Musical Excerpts, Chords and Scales Using a Diverse Array of Hardware." ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2025.

@inproceedings{pedroza2025guitar,
  title={Guitar-TECHS: An Electric Guitar Dataset Covering Techniques, Musical Excerpts, Chords and Scales Using a Diverse Array of Hardware},
  author={Pedroza, Hegel and Abreu, Wallace and Corey, Ryan M and Roman, Iran R},
  booktitle={ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={1--5},
  year={2025},
  organization={IEEE}
}

Guitar-TECHS

Guitar-TECHS is a comprehensive electric guitar dataset designed to advance machine listening research in tasks such as timbre transfer, performance generation, and automatic transcription. It captures performances by three professional guitarists (P1, P2, P3) across diverse recording environments, hardware setups (guitars, amplifiers, microphones), and musical content. The dataset includes multi-perspective audio signals (direct input, amplifier microphone, egocentric/exocentric stereo microphones) and synchronized MIDI annotations per string, enabling robust model training and analysis.

Key Features

• Diverse Performers & Gear: Recorded by three professional players in distinct rooms, using varied guitars, amplifiers, microphones, and audio interfaces.

• Multi-Signal Capture:

  • Direct input (DI)

  • Amplifier microphone

  • Egocentric (head-mounted stereo microphone)

  • Exocentric (front-facing stereo microphone, 5 feet from performer)

• Musical Content:

  • Techniques: Vibrato, harmonics, bendings, palm mute, pinch harmonics.

  • Scales: Played in "boxes" at 120 BPM.

  • Chords: 20+ chord types across 5 sets, with inversions (see Dataset Structure).

  • Musical excerpts and single-note recordings.

• MIDI Ground Truth: Per-string MIDI annotations via Fishman Triple Play Connect pickup.

(note: the ego and exo signals were captured with a video device, but Guitar-TECHS only includes this device's audio capture)

Dataset Structure

Track Categories

1. Music: Full musical excerpts.

2. Single Notes: Individual notes.

3. Scales: Played at 120 BPM.

4. Techniques: Articulations (e.g., bends, palm mutes).

5. Chords:

   • Set1: Strings 1,2,3 (maj, min, aug, dim) | Strings 1,2,3,4 (Maj7, 7, m7, m7b5).

   • Set2: Strings 2,3,4 (maj, min, aug, dim) | Strings 2,3,4,5 (Maj7, 7, m7, m7b5).

   • Set3: Strings 3,4,5 (maj, min, aug, dim) | Strings 3,4,5 (dim, aug).

   • Set4: Strings 4,5,6 (maj, min, dim, aug).

   • Drop3: Strings 2,3,4,6 (Maj7, 7, m7, m7b5).

   • Chord types include: Major, minor, augmented, diminished, Maj7, dominant 7, minor 7, minor 7♭5.

Note on temporal aligment: Signals may exhibit ≤100ms misalignment due to recording setup. For alignment-critical applications, apply correction (e.g., in the __getitem__ method of a PyTorch Dataset subclass).

Website: guitar-techs.github.io (includes paper and further hardware specifications).

Other related works: 

Pedroza, Hegel, et al. "EGFxSet: Electric guitar tones processed through real effects of distortion, modulation, delay and reverb." ISMIR (2022) https://egfxset.github.io

Pedroza, Hegel, et al. "Leveraging Real Electric Guitar Tones and Effects to Improve Robustness in Guitar Tablature Transcription Modeling." DAFx (2024) https://robust-guitar-tabs.github.io

Pedroza, Hegel, et al. "EGSet12: twelve real & original solo electric guitar performances with diverse playing styles to evaluate guitar tablature transcription." Zenodo (2024). https://zenodo.org/records/11406378