The biggest music labels are already dealing with the challenges posed by the application of artificial intelligence technologies. But this is just the beginning, because scientists from Claremont-Gradwaite University (USA) have taken it to a new level by teaching AI to predict potential hits.
In the process of training the system, the researchers used the neural activity of 33 volunteers aged 18 to 57, who were asked to listen to 24 musical compositions selected by a special streaming service.
The data obtained was processed using a statistical model, which made it possible to train artificial intelligence to predict potential hits.
The selection included songs of various genres, including 13 hits and 11 songs that were not successful. A hit status was assigned to a song that received more than 700,000 streams on the platform.
At the end of the listening, each participant filled out questionnaires in which they had to indicate their opinion about the compositions they listened to. The study organizers were interested in whether the participants found the compositions offensive, whether they had heard them before, and whether they could recommend them to their friends.
The main aspect of the study was the natural psychophysical reaction of the listeners to the compositions. Participants were divided into samples of 33 people, and 24 songs were listened to. This turned out to be enough for the accuracy of predicting the results of the study to be at a high level.
Using a linear statistical model, results were obtained that made it possible to estimate the probability of success in determining a potential hit at 69%, writes TASS .
However, the artificial intelligence algorithm showed higher accuracy and gave a result with a probability of 97.2%. In the course of the complication of the task, it was enough to listen to the composition for only one minute to get the results.
Read the Latest World News Today on The Eastern Herald.
