Tag Archives: musicology

Chordify team members publish an article about subjectivity in chord recognition

Chordify’s best and brightest are constantly researching new ways to improve our chord recognition algorithms. They work in a field called music information retrieval (MIR): which is a scientific area that combines disciplines like musicology, artificial intelligence, and informatics.

Recently published in Journal of New Music Research, the article “Annotator subjectivity in harmony annotations of popular music,” questions whether experts listening to music can agree over the chords they hear. The article came out of a collaboration with researchers at the Utrecht University, the University of Amsterdam and the University of Southhampton. 

Added harmonies

The problem of recognizing which chords are played in a musical recording is not new. For instance, when hearing back their song “A Hard Day’s Night,” Paul McCartney, George Harrison, and Gary Moore (of the Beatles) couldn’t agree over the opening chords. The article puts forth that this ambiguity may be due to the sounds added by electronic production techniques. In other words, your phasers, distortions, flangers, and choruses are adding sounds that create new harmonies, and thus make new chords. 

Number crunching

Scientists wouldn’t be scientists if they didn’t do some number crunching. So here we go. The dataset, called the “Chordify Annotator Subjectivity Dataset” (CASD) consists of 50 songs, which are all provided with chord labels by 4 musicians. These annotators have between 15 and 25 years of experience, and are highly educated listeners. They took 20 minutes and 25 seconds on average to transcribe a song; and they rated the difficulty of the songs with a 2.1 average, on a scale from 1 to 5 (where 5 is the most difficult). Finally, in total 290 unique chords were used. If you want to read the full scope of the study, you can get the article here.  

No general truth

In conclusion, the article states that the musicians in the study agreed only 76 percent of the time, over simple chords. And that’s not even talking about complex experimental jazz harmonies. When it comes to those, the level of agreement went down to a mere 59 percent. So they argue that the idea of a general ‘truth’ is hard to keep up, even when it comes to chords. Happy jamming!