Engineers have a significant tendency to being nerds. And nerds attract nerds. Meaning — we have an office full of nerds! Hooray :-) One of the many upsides with this is our dedication to very important things. Such as building apps deciding whether a song is Pop or Rock. Seemingly absolutely pointless. But wait — is it, really?
Nerd is our middle name. And Passion. And definitely Pop. Or, was it Rock?
For you restless people who want to cut to the chase: We built an AI app with BERT (Bidirectional Encoder Representations from Transformers) that lets you fact-check your music knowledge. Try for yourself — is your music intuition as good as you think? And here’s how you can build the model yourself.
Many of us at Peltarion are true music nerds, so music is a topic where we easily get stuck in endless discussions. And stuck is the point here, not only in arguments, but in processes, businesses, and life. This time, we got stuck in a discussion about whether a song was Pop or Rock? A deadly serious discussion! Our biggest talent for diplomacy, Daniel, also Head of Engineering, decided to solve the big issue and came up with the idea of building an AI app to sort out whether this song — or any other song for that matter — was, in fact, Pop, or, Rock. Here’s the app.
Here's how we got unstuck
Somehow, Daniel found a music dataset that was publically available that he could upload to our platform, and the data he chose was labeled with “pop” or “rock” (we dismissed hip-hop, this time). To get good results when analyzing the text he used the AI model BERT, and after a good day of training, he got a result with decent accuracy. As Daniel is a bit of a multi magician he did a quick design and deployed the model to an app. All in all, if you would copy-paste his procedure, building a model like this will take about 30 min (+8h of training) (tips: while the model is training, get yourself a good night's sleep ;-).
The Pop or Rock app was a lifesaver for our history of music battles and we ended the next discussion as friends, almost...
But uhm — is this really useful?
Not really, no. But -
The serious business case here is the automation of manual classification. The time (& money) AI can save on a manual task like that has huge potential. The same principles as in the Pop or Rock app can be replicated for tasks where understanding the meaning of words is needed — regardless of industry.
Copy that & build it yourself
To do this yourself, it turns out you don’t need to be a super engineer or a data scientist. Thanks to our tech writer, Calle, who explained how to do this in a tutorial, you can follow his steps and build it yourself!