Xavier the Spotify AI DJ

Spotify recently joined the AI hype by introducing a new AI DJ to their app. I was initially deeply sceptical and cynical and started muttering about cancelling my subscription and just going back to using my local library and Plex. However, I thought I’d give it a go. So far I’ve found it to be… not terrible.

The Good

The text-to-speech is pretty nice. Spotify hired an actual human, Xavier Jernigan to provide the training data for the voice model and it feels like it was probably a worthwhile investment. Human Xavier’s voice is perfect for DJing and much less jarring than I imagine the default Siri or Google voice would be.
It’s kind of weirdly nice having a ‘DJ’ pop up occasionally between songs and give some facts about what I’m listening to. Xavier sometimes gives little factoids about artists and albums that he’s introducing.
I quite like the fact that he changes up the set every 5-or-so songs. Previously I just had a giant playlist that I share with my wife with everything we both like on it and we just played that on shuffle so it’s kind of fun to have a few minutes devoted to a certain ‘vibe’ or artist without having to manually make that happen.
Sometimes Xavier plays new songs, so it’s quite nice from a discovery angle – I’ve found a few songs that I’d not heard before that have gone onto my giant list of songs including Fairlies by Grian Chatten and Nothing Matters by The Last Dinner Party.

The bad and ugly

If you listen to Xavier for a few days in a row, it does start to repeat stuff quite a lot. Not necessarily a problem if the repeated material is stuff you like but a little jarring if prefixed with “here are some new songs you might not have heard before” – yes actually, you played them for me yesterday!
Genre names seem to be a bit weird, I assume that’s actually a symptom of how Spotify tags up its music rather than a problem with the AI but again, it’s a little jarring when Xavier says “Now it’s time for some indie hits”, your mind jumps to the mid-00s and prepares for Arctic Monkeys or Pigeon Detectives and you actually get a Blur song (I love Blur by the way, but they are Britpop, not indie).
Sometimes Xavier seems to play 5 songs by an artist I can’t stand. I’m not quite sure why that might be but at least you can skip through the set.
There seem to be a few places where Xavier doesn’t work yet including Spotify’s desktop app and if you stream via Alexa. I imagine they’ll fix that. You can still listen to “AI DJ” playlists over Alexa but you just don’t hear the voice-overs and thus don’t have any context for each set.

Conclusion

I’ve quite enjoyed experimenting with Xavier. I hope that the functionality improves and that the repetitive behaviour can be fixed. I’ll keep listening to Xavier until I get to the point that the suggestions irritate me a bit too much.

Also as a geek, I’m also interested in whether an open/alternative version could be cobbled together using open metadata from places like MusicBrainz and Last.fm and even libre.fm scrobbles, a text to speech engine, a free Llama-like LLM and a sprinkling of data science to cluster similar songs and artists.