Visual Pitch Class Profile - A Video-based Method for Real-time Guitar Chord Identification

We propose a video-based method for guitar chords identification which is analogous to the state-of-theart audio-based method. While the method based on audio data uses the Pitch Class Profile feature and supervised Machine Learning techniques to “teach” the machine about the chord “shape”, we use as feature the approximated positions of fingertips in the guitar fretboard, captured using especial hardware. We show that visual and audio-based methods have similar classification performance, but the former outperforms the later with respect to the immunity to noise caused by strumming.