Idea: Procedurally Generated DJs

I enjoy listening to what we now call “radio” on services like Pandora and Spotify, where the songs are selected for me based on my preferences. But sometimes I miss the voice of a DJ telling me what I heard.

It seems that with all the text-to-speech technology we have today, it shouldn’t be too hard to generate a DJ on the fly. A realistic voice could come in occasionally as a song fades out and recap the last few tracks for me. Maybe it could even tap into an API to give me information about upcoming concerts the way a local DJ would. And then it could do that thing where it cues up the next song and stops talking right before the lyrics start.

It might sound something like this:

And as long as we’re hooking into APIs, how about adding some news, traffic, and weather? And since it’s all specific to your account, it could look at your actual calendar to give you personalized information. You could even have a second voice as the news-and-weather sidekick.

That might sound something like this:

Maybe you have a favorite DJ who’s not around anymore. Maybe you really miss Casey Kasem or Wolfman Jack. Thanks to the many hours of audio we have of them, surely we could train a text-to-speech generator to sound just like them, right?

Of course, that wouldn’t capture their personality. To approximate that, we’d need to turn to artificial intelligence.

I took an actual quote from Casey Kasem and fed it into a neural network to see if it could generate some convincing DJ patter. In this quote, the bold text is what I provided, and everything else afterwards is generated by artificial intelligence:

That’s coherent, I guess, even if it’s completely inaccurate. Also, I just went and listened to “The Chair” and it’s not a song I want to hear again. Sorry, George. It’s just not for me.

Let’s try that again with the same prompt and see what else the neural network comes up with:

That’s ridiculous and amazing. I would totally listen to this robot DJ.

I’ve picked a random Iggy Pop song to try out with that last paragraph to see how it sounds. It works pretty well (although this is not meant to be simulating Casey Kasem’s voice):

Now let’s try to generate a robot Wolfman Jack. Again, the part in bold is what I fed the neural network based on actual Wolfman Jack patter, and the rest is all generated by artificial intelligence:

Uh. The sentences make sense but that takes a left turn pretty quickly. Maybe one more try. Same prompt:

Wow, that’s a lot better. Yeah, I’d listen to robot Wolfman Jack.

So maybe a procedurally generated DJ with personality is still a bit hard to make. But I think a basic A.I.-powered DJ can’t be far away.

And maybe it can play some A.I.-generated music in the style of your favorite artists, like this Beatles soundalike album generated by OpenAI Jukebox:

This Exists! Mostly.

Before I wrote the above, I did a Google Search to see if this had been done yet, and found a lot written about how A.I. can handle the music-selection part of being a DJ, but nothing about the patter. Two days after I wrote it — one day before this newsletter was sent out — I modified my search terms and actually found something that’s very much a step in this direction! (Great minds, etc.)

It’s called Radiant, and it’s been developed into a viable product powered by Spotify. There’s no neural-network-derived personality, but they did give their DJ “Rad” a snarky robot personality of his own. I kind of love it, even though it needs some work. In my tests, Rad repeated some patter, played the same PJ Harvey song three times, and was drowned out by Fiona Apple. And there are typos in the interface. I’m not sure I’m ready to make Rad my full-time DJ yet, but it’s a great first step that I’ll be playing with and keeping an eye on.

Judging by the six reviews on the app store, and only 76 followers on their dormant twitter account, it looks like Radiant hasn’t gained much traction. But it’s a free app. Go download it! Maybe we can get them the traction they need to keep developing it further!

The developers wrote this piece on how they developed Radiant, and here’s their YouTube demo of the product:

Don’t Forget To Ring The Bell

If this were a YouTube video, this is where I’d remind you to subscribe, like, and click the little bell. If it were a podcast, this is where I’d ask you to rate the podcast on iTunes. Those are things that help those platform’s algorithms figure out what to recommend to you and others.

But Substack has no such algorithm. As they recently said in a blog post* about content moderation:

That’s generally a good approach for the platform, I think. But it’s a double-edged sword. As a reader, how do I discover good newsletters? Substack has some discoverability features in their new web-based newsletter reader (which must use some sort of algorithm to sort its recommendations, right?) but it only includes newsletters from their own platform.

Substack wants to keep you reading on Substack. And YouTube’s algorithms are designed to keep you watching on YouTube. But iTunes hosts podcasts from all over, and that’s where people go to rate podcasts. So where’s the decentralized review center for newsletters?

Of course, email newsletters have been around for decades and never needed anything like this before (or maybe I just never looked for it before). But it feels like they’re having a moment as the latest content bandwagon, so maybe the time is now. Does something exist like Goodreads for newsletters? Are there any media outlets with critics covering the Newsletter beat?

If you have a favorite way of discovering newsletters, let me know.

In the meantime, here’s a bell you can ring to share this one: 🔔

*Yes, Substack, the platform for newsletters that’s rapidly replacing blogs, has a newsletter called the Substack Blog that is available as both a newsletter and a blog. The lines they are a-blurring.

That’s it for today. I apologize to real DJs with deep understand of music history and jobs that will never be truly replaceable by robots. There are so many DJs who introduced me to music that profoundly influenced me, including Bruce Kelly, the Bone Mama, Vin Scelsa, Megless Griffin, Doctor Demento, tons of people I’m forgetting, and everyone who influenced them.

See you next week! Until then, keep your feet on the ground and keep reaching for the stars, and remember the words of Robot Wolfman Jack: “Life is pain. But if you eat enough pills, you can deal with it.”

David