Predicting which songs will win prestigious awards like the Grammys has long been a mix of art and speculation.
But now, researchers at New York University have developed an artificial intelligence (AI) tool that takes the guesswork out of this process by analyzing specific traits of Grammy-winning songs.
Their work offers new insights into what makes a song successful and could even help discover emerging artists and trends.
The study, led by Anasse Bari, a clinical associate professor at NYU’s Courant Institute of Mathematical Sciences, focuses on three major Grammy categories: Song of the Year, Record of the Year, and Rap Song of the Year.
The researchers wanted to understand not just which songs might win, but also what qualities these songs share.
“Spotting award-winning art is a subjective process, complicated by the secrecy of voters’ decisions,” Bari explains. “However, by analyzing what we know about the songs—their composition and popularity—we can pinpoint those likely to win.”
To build the AI tool, the researchers created a dataset of nearly 250 songs nominated for these awards between 2004 and 2020. They then trained AI algorithms using various factors, including Billboard rankings, Google search volume, and specific musical characteristics provided by Spotify data.
Some of the key musical features analyzed by the AI include:
- Acousticness: Whether the track relies on non-electric instruments or sounds
- Danceability: How suitable the track is for dancing
- Energy: A measure of the song’s intensity and activity
- Instrumentalness: The presence or absence of vocals in the track
- Speechiness: The amount of spoken words in the song
In addition to these musical traits, the researchers also analyzed the lyrics of each song using Natural Language Processing (NLP) algorithms. These algorithms examined the diversity of vocabulary, emotional tone (happy, sad, angry), and the use of profanity in the lyrics.
The AI then generated a list of the top three candidates for Song of the Year, Record of the Year, and Rap Song of the Year for each of the years studied (2021-2023). Remarkably, the AI tool successfully included all nine Grammy-winning songs in its top three lists. These included Billie Eilish’s “everything i wanted” (2021 Record of the Year), Silk Sonic’s “Leave the Door Open” (2022 Song of the Year), and Kendrick Lamar’s “The Heart Part 5” (2023 Rap Song of the Year).
Interestingly, the AI’s predictions sometimes differed from those made by betting sites. For example, Bonnie Raitt’s “Just Like That,” which won the 2023 Song of the Year, was ranked among the top three by the AI tool but was considered unlikely to win by gambling platforms. Similarly, H.E.R.’s “I Can’t Breathe” was predicted by the AI to be a strong contender for the 2021 Song of the Year, even though it was viewed as a long shot by betting sites.
The study also revealed that different musical features were more predictive depending on the award category. For Song of the Year, factors like energy, acousticness, and peak Billboard position were key.
For Record of the Year, speechiness, profanity, and acousticness were more important. For Rap Song of the Year, the number of words, vocabulary diversity, and happiness score were significant predictors.
While the researchers caution that their AI tool is not a precise prediction device, it does highlight the complex mix of factors that contribute to a song’s success. As Bari notes, “Our findings show the potential of using machine learning and data-driven techniques to gain insights into what makes a song a hit.”
This AI tool not only offers a new way to predict Grammy winners but could also help in identifying future musical trends and discovering artists who might otherwise go unnoticed.
