34  How Capo Detects Chords

Capo contains a sophisticated neural network model that can estimate the chords by “listening” to the music stored in audio files. This network was trained using a large collection of songs that, when played one after another, would take more than a whole month to listen to.

The result of this training is a very accurate and capable recognition engine that goes beyond the basic major and minor chords. In total, Capo can recognize 14 different chord qualities including major 7, minor 7, suspended, and augmented chords. Additionally, Capo can also spot inverted chords (e.g., D/F#).

34.1 The Limits of Capo’s Chord Detection

Like all things of this nature, Capo’s chord detection isn’t perfect and will make mistakes at times. This is why Capo offers a full suite of tools for modifying the estimated chord results.

Songs that don’t have very strong (or mostly constant) harmony might not produce very many chords, or none at all. In these cases, you can apply your own harmony (say, to build a cover version) by placing your own chord entries using the chords controls or k key.

The worst offenders are songs that are horribly out of tune—more than a quarter tone away from the A440 reference tuning. With these tracks, you’ll notice two major problems: (1) the chords are a semitone away from the true pitch (e.g., a C chord will be recognized as B), and (2) the audible feedback sounds horribly out of tune when you select chords. This can be fixed by exporting a pitch-corrected version of the song, and creating a new project using that audio.