“The problem with our industry,” moaned the record executive, “is that we’re always working blind. We’re totally at the mercy of a fickle public. We think we might know what they want, but when we release a single, we have no idea whether it will catch on. You can throw all the money in the world at a song, but if the public doesn’t bite, it all goes down the toilet.”
He wasn’t wrong. Sure things in the music business don’t exist. But labels can embark on all kinds of research to mitigate mistakes and dead ends. Increasingly, that means turning to algorithms that can predict if a song will be a hit.
Many companies use a program called Hit Song Science, an AI method of analyzing a song for hit attributes. By comparing data points derived from thousands of hits going back to the 50s to a new song, it can offer an opinion on the potential popularity of new song.
The latest entry in this arena is Hyperlive, a startup from Finland that says its algorithm is 84% accurate when predicting hits. From Digital Music News:
Hyperlive’s algorithm correctly predicted how 10 tracks from major artists, including Ed Sheeran and Taylor Swift, would do. Since their release, the songs have accrued a combined 180+ billion streams and over 1.2 billion single sales.
Analyzing each track’s audio signature, Hyperlive’s algorithm predicted actual performance with 84% overall accuracy.
In addition, for tracks incorrectly identified as hits, predicted total track sales and streams fell within an average of 25% of their actual range.
From a business point of view, I get it. Labels need to reduce the risk when it comes to releasing music. But won’t this just created more and more songs that sound more and more alike?
Read more here.