AI’s Singing Talent is Improving, But its yet Not Ready to Topple Taylor Swift

Let's explore AI's singing talent through the cover of Dolly Parton's "Jolene by Holly Herndon

As machine-learning music specialist Prof Nick Bryan-Kinns explains, new neural networks are capable of writing original music – but may never compose meaningful lyrics. Take a hike, Bieber. Step aside, Gaga. And watch out, Sheeran. Artificial intelligence is here and it's coming for your jobs. That's, at least, what you might think after considering the ever-growing sophistication of AI-generated music. While the concept of machine-composed music has been around since the 1800s (computing pioneer Ada Lovelace was one of the first to write about the topic), the fantasy has become reality in the past decade, with musicians such as Francois Pachet creating entire albums co-written by AI. Holly Herndon has released a new cover of Dolly Parton's "Jolene", but with a twist: the single was recorded with artificial intelligence. Let's explore AI's singing talent.

The AI cover was created using Herndon's deepfake twin, Holly+, who was able to sing the iconic song in Herndon's voice. "I thought it would be fun to share the latest version of the Holly+ voice, which is so realistic that it seems some people haven't realised it is entirely AI-generated," Herndon tells Dazed.

"It felt interesting to cover Dolly for a few reasons. I've been exploring the idea of Identity Play (IP), or using AI to be able to perform with or as someone else, and Dolly's music means a lot to me as we are both from Appalachia. This is the kind of song I would never dare to sing with my natural voice. She is a fantasy figure for me," she continues.

The track has been released alongside a new music video, directed by digital artist Sam Rolfes. For the video, Rolfes used motion-capture technology to animate a 3D model of Herndon. "My friend Sam Rolfes is my favourite mixed reality artist, he is a real virtuoso. We discussed these ideas of pastiche and identity play, and we decided he would perform as me performing as Dolly wearing rhinestone nudie suits that were a fantastical representation of Appalachia when I was growing up," she says.

But the performance is also a little unsettling. For one, the giant inhales between verses are too long to be real and are almost cajolingly dramatic. The vocals themselves are strangely even and, despite the somber tone affected by the AI, lack Parton's iconic vulnerability.

Overall, it feels like the AI is simply checking the boxes of what makes a good, swooning cover after listening to Jeff Buckley's "Hallelujah" a million times — which, to be fair, is a pretty good starting point.

Still, it'd be remiss to downplay what Herndon has managed to pull off here, and the criticisms mostly reflect the AI's limited capabilities more than her chops as a musician. The AI's seams are likely intentional if her previous work is anything to go off of.

Either way, if you didn't know you were listening to an AI from the get-go, you'd probably be fooled. And that alone is striking.