Listen to an AI Deliver an Uncannily Human Performance of 'Jolene'

AI-powered image generators have recently dominated headlines, but musical machine learning models have made significant advancements in their own right. At the forefront of this evolution is Holly Herndon, who, along with partner Mat Dryhurst, co-developed Spawn—a singing neural network featured on her last album, Proto. Last year, they released Holly+, allowing users to access a model of Herndon's unique voice.

Herndon has now unveiled a new single, a cover of Dolly Parton's "Jolene," performed entirely by her digital twin. At first listen, the track may come off as slower and quieter, with Ryan Norris providing the instrumentation without straying too far from the original arrangement. The essence of the song shifts from frantic desperation to a more plaintive resignation.

What makes the rendition truly captivating is that every vocal nuance, down to the sharp inhales before the harmonies, was generated by Holly+. Remarkably, there isn’t a human vocalist involved; the model replicates Herndon's voice remarkably well despite some occasional stilted phrasing and digital artifacts upon close inspection.

Until now, many top artists exploring AI have tended to focus on creating generative soundscapes or synth melodies. This single marks a pioneering moment, as it is, to my knowledge, the first instance of a machine learning model taking the lead vocals in a pop song.

Herndon initially showcased the track at Sonar Festival in March, but it flew under the radar until its official release this week. For an exciting preview, the Sonar presentation features fascinating real-time demonstrations of Holly+ and the innovative technology from Never Before Heard Sounds.

If you're interested in experimenting, you can record your own rendition of "Jolene" and upload it to Holly+, though the fidelity may not match the original track. For artists serious about incorporating AI into their work, Herndon and Dryhurst have launched Spawning, an organization dedicated to advancing AI creativity in music.