For Spotify, the creation of its artificial intelligence DJ — which launched in beta version for Premium subscribers in the U.S. and Canada on Feb. 22 — represents the culmination of what the company has been working toward for years.
“This has always been a north star,” Ziad Sultan, the vice president of personalization at Spotify, tells PvNew. The DJ serves as a curator, analyzing user activity to play songs that it knows users already like and offering recommendations for new music. It also provides commentary and background context on the tracks and artists it plays. Playlists like Discover Weekly, which feeds recommendations to users based on their listening activity, or Spotify Wrapped, which pulls together users’ most-streamed songs at the end of the year, mimic some of the AI DJ’s functionality, but the newly debuted product serves as the embodiment for what Sultan says had been a “metaphor” for the company until now.
The process of creating the DJ was contingent on pulling together several major pieces. The company’s personalization technology, plus an AI text-to-speech engine from recent acquisition Sonantic and technology from OpenAI, provided the basis for the product. Add the voice of Xavier “X” Jernigan, Spotify’s head of cultural partnerships, plus the work of a writers’ room headed by Sulinna Ong, and the AI DJ was born.
Jernigan, a music industry veteran who worked in marketing for Universal, Sony and Def Jam before joining Spotify in 2016, previously served as host for Spotify podcasts “Showstopper” and “The Get Up.” He broke down the training for the AI DJ:“I had directors in the sessions with me, just to make sure I’m sounding like me. Then they would come up with things that we know we needed to focus on to train the model.” Additionally, his voice clips were stripped from “The Get Up” and then uploaded into the model.
With the DJ, he wanted to tap into several different characteristics: “The things we talked about are friendly. Relatable, like your buddy. Knowledgeable, but not arrogant.”
The process of creating the DJ also involved putting together a weekly writers’ room to script what the DJ would say. Ong, the global head of editorial, says that the writers spend time thinking about topical subjects, from calendar moments to cultural sensations based on pop culture. “The listening trends or habits are quite predictable,” she says. “Whether that’s Christmas or Valentine’s Day, and you think about thematic and calendar moments like that. Let’s say, if we had ‘Stranger Things’ airing, or ‘The Last of Us,’ for example at the moment, we know that there’s going to be interest in those songs. Those are the things that we discuss in the writers’ room — topical moments.”
Another factor is honing the DJ’s voice, which is largely based on Jernigan’s conversational style. “When you think about AI voices, you think something that’s going to be quite robotic,” she says. “X really has a personality in his voice that you can hear. When we’re writing things in the writers’ room, we first of all think about the context and the cultural relevance. Then X will come in and tinker it to make sure that it’s in his own voice.”
In the future, Ong hopes Spotify looks into the possibility of adding multiple voices.
The AI DJ had a spotlight moment in Spotify’s Stream On event on Mar. 8, which is when the company debuted its new user interface and announced exclusive new partnerships. According to company data from Feb. 22 to March 1, users who can access the DJ have devoted a quarter of their listening time to it, with over half of first-time listeners using the product again the next day.
VIP+ Analysis: What AI DJ Can Teach Video Streamers
