A brief note on Spotify’s AI DJ
I had anticipated this article being much longer, as I delved into the murky data attribution practices behind Spotify’s AI DJ. The theft of data from marginalised creators is so commonplace, I was expecting to find that Spotify had scraped recordings from actual black DJs from radio recordings and podcasts hosted on their site to create a synthetic DJ with a blaccent. However, I was pleasently surprised to find that the voice actor is actually named (Xavier “X” Jernigan, a senior member of Spotify’s staff), and (presumably) consented1.
The AI DJ uses a text-to-speech model from Spotify-acquired company Sonantic, trained on “a few hours”2 of Jernigan’s voice. Sonantic’s original financial model compensated the voice actors for every use of the model2, and we can only hope Jernigan is well compensated for the use of his voice in the original Spotify AI DJ model. Let’s hope also that as Spotify “continue to iterate and innovate”1 they’ll be sure to gather data in a transparent and consent-driven way.