Nadar, Christon-Ragava; Abesser, Jakob; Grollmisch Sascha

Towards CNN-based acoustic modeling of seventh chords for automatic chord recognition, Proceedings of the 16th International Conference on Sound And Music Computing (SMC), Málaga, Spain, 2019.

Dataset Overview

The IDMT-SMT-CHORDS comprises of 16 MIDI generated audio files consists of various chord classes. Here we focused on chord voicings, which are commonly used on keyboard instruments and guitars. Based on this we categorized as Guitar and Non-Guitar instruments. We used several software instruments from Ableton Live and Garage Band to synthesize these MIDI files with various instruments such as piano, synthesizer pad, as well as acoustic and electric guitar.

  • File duration: 4.1 Hours
  • # Chord segments: 7398
  • # WAV files: 16
  • Chord duration: 2 seconds
  • BPM: 120
  • Time signature: 4/4
  • Sampling rate: 44.1KHz
  • Mono audio


The Non-Guitar files includes all chord types in all possible root note positions and inversions. For example, C Major triad chord is included with its two possible inversions C/E and C/G.

All non-guitar chord classes are listed below:

  • Major (+ 2 inversions)
  • Minor (+ 2 inversions)
  • Major 7 (+ 3 inversions)
  • Minor 7 (+ 3 inversions)
  • Power Chord  - root and fifth note (+ 1 inversion)
  • Dominant 7 (+ 3 inversions)
  • Minor 7 flat 5 (+ 3 inversions)

This gives us 576 non-guitar chord classes.


The guitar files where generated based on barŕe chord voicings with the root note located on the low E, A, and D strings. For example, to modeling major chord and it’s voicings we use open position E maj, A maj and  D maj shape and move 12 steps (including octave at 12th fret) thereby we get 39 positions (13*3).

List of  Guitar chord types:

  • Major (+ 2 voicings)
  • Minor (+ 2 voicings)
  • Major 7 (+ 2 voicings)
  • Minor 7 (+ 2 voicings)
  • Power Chord  - root and fifth note (+ 2 voicings)
  • Dominant 7 (+ 2 voicings)
  • Minor 7 flat 5 (+ 2 voicings)

This gives us 273 different guitar chord classes.

More audio datasets

Fraunhofer IDMT has compiled further audio data sets for various research areas such as instrument recognition, fingerings or game analysis.

Acoustic methods for quality assurance in industrial use

Audiovisual content analysis and trusted AI solutions for media