In the past you had to go over record heads to get around your song, nowadays you just toss that potential hit online to see where the ship’s leads go. As a result, tens of thousands of songs are released every day. But how do you identify gems? Here too, technology offers a solution.
It’s hard for radio stations and streaming services to separate the wheat from the chaff, when of course that’s their job: they have to deliver what you want to hear. To choose, they allow people to listen to all the music. They then have to evaluate which songs will appeal to a large audience. But even with the help of artificial intelligence, it was not possible to correctly predict more than 50 percent of the results. Big waste of time and effort. And listeners are often presented with music they can’t wait for.
Pretty much perfect
American researchers believe that this could be even better. to New machine learning technology When applied to brain responses, they were better able to predict which song would be a hit. In fact, they got it right 97 percent of the time. Paul Zak, Professor at Claremont Graduate University. “The neural activity of just 33 people who can predict whether millions will listen to a particular song is really special. Nothing comes close to this high accuracy.”
The brain betrays preference
How did that work then? Participants were given special sensors while listening to 24 songs. Then they had to define their preferences. During the experiment, the scientists measured the neurophysiological responses in the participants’ brains to the songs. “The brain signals we collected showed the activity of the brain network that controls mood and energy levels,” Zak explains. Based on this, researchers can predict which song will conquer the world and which will not. So this method measures neural activity in a small group of people in order to make population-level predictions without having to record the brain activity of hundreds of people.
After collecting the data, the researchers first ran some statistical analyzes themselves, but to improve the predictive power of the method, they then trained a machine learning model, which tested different algorithms to come up with the best predictions.
Better playlist
And it helped: the statistical-only model had a 69 percent success rate, while machine learning correctly predicted the outcome 97 percent of the time. It could be faster. If they only apply machine learning to the first minute of a song, the accuracy is still 82 percent. “This means that streaming services can easily identify new songs that are likely to become hits. This makes their job easier and the listener gets a better playlist,” says Zak.
The future in that region is much brighter. “When the wearable technology we used in this study becomes very popular, the right entertainment can be sent directly to the right audience, based on their neurophysiology. Instead of being offered hundreds of choices, you get two or three more. This makes it easier to choose the music you like best.” Much and faster,” says the scientist.
Also for TV
But we’re not there yet. The researchers also mentioned some caveats to their study. For example, they used relatively few songs. The question is how correct is the machine learning model when adding hundreds of songs. However, they are confident that their method will be widely used and not just for music. “Our most important contribution is methodology. This method can potentially be used in all kinds of entertainment, such as movies and TV series.
