About a month ago, Chordify’s own Bas de Haas and Vincent Koops flew down to Suzhou, China, to visit the annual Conference of the International Society for Music Information Retrieval (ISMIR). Not just to check out the latest trends and research, but to contribute scientifically, by sharing a large open data set. ISMIR is the world’s leading research forum on processing, searching, organizing and accessing music-related data.
Co-founder Bas is a real veteran when it comes to the conference, having been involved with music information retrieval ever since his PhD research. In fact, the very first algorithm that automatically extracted the chords of your favorite songs is based on his academic handiwork.
Measuring subjectivity
The funny thing about finding chords in songs, is that it’s not as objective as you might think. “Musicians sometimes disagree on the chords they hear”, Bas explains. “That’s why you’ll find a lot of different versions of tabs and sheet music of one song on Ultimate Guitar, for example. Simply put, depending on your level of experience, or even the instrument you play, musicians don’t always agree on what the ‘right’ chords are in a song” “
This is something we call Annotator Subjectivity and up until now, it wasn’t something that you could really measure”, Bas continues. “Together with the Utrecht University, where Vincent is doing his PhD research, we wanted to see if this was actually something you could measure. Vincent did the bulk of the research and we asked four professional musicians, two guitar players and two pianists, to write down the chords of 50 different pop songs. And guess what? Even though they all are very skilled and experienced, they only agreed on about 80% of the most basic chords.”
Why is that relevant? Chordify automatically extracts the chords of a song of course, so why is this research useful? The cool thing about this, is that it’s not just about the right or wrong chord. Different musicians would agree that a chord is an A minor for example, but they can disagree on whether there’s an added 7th or augmented 5th.
That sounds difficult to understand, but it shows that pianists and guitarists listen to chords differently. “Chordify could apply this research in very cool ways”, Bas says. “In the future, we could give people customized chords for their instruments, for example. But we could also customize things to suit your level of playing. If you’re a beginner, you’ll get easier versions of the chords, and if you’re getting better, you’ll get more complex versions of chords.” “We also think it’s important to share this kind of research, so that’s why we released an open data set. Not just for the ISMIR conference, but for anyone interested.” We have our own GitHub page, so be sure to check it out! https://github.com/chordify/CASD