James Cridland’s International Radio Trends: Robot radio comes closer, with RadioGPT – but is it any good?

James Cridland, radio futurologist, is a conference speaker, writer and consultant. He runs the media information website media.info and helps organise the yearly Next Radio conference. He also publishes podnews.net, a daily briefing on podcasting and on-demand, and writes a weekly international radio trends newsletter, at james.crid.land.

A press release from Futuri Media promised a thing called RadioGPT™. It read like the sort of thing that I should be implacably against.

“The World’s First AI-Driven Localized Radio Content”, RadioGPT is a system that a) watches social media for trending topics in your local area; b) writes a script apparently using GPT-3 technology; c) uses AI voices to make radio with them. It has the horrible title of “live, local, and powered by AI”.

I shared the press release on social media. People responded with the usual “this sounds like an awful idea. AI voices aren’t there yet. Nobody will use this. This is the worst idea”. And I hoped that was true.

But then, I discovered that there was a live stream demo of this dreadful service that anyone can listen to. So, I took a listen (while on a train in Sydney), looking forward to being really rather rude about this horrible idea.

Irritatingly, though: it’s rather good.

Now, there are some smoke and mirrors being deployed here by Futuri for the test stream. It isn’t live: it’s a loop. The voices may be AI, but either they’re the best AI voices out there, or someone’s tuned their delivery rather suspiciously well. I know that TopicPulse does use some human moderation, too. The stream cleverly just says every voice is AI: this isn’t a 100% AI-produced service.

There are some clever tricks, too. The songs in the demo stream are all 90-second edits (good for a demo). The local news is for “Springfield” – someone’s a fan of the Simpsons at Futuri towers.

The little bits of news, scripted to have a little joke at the end of them, work well. The station sounds tight. It’s undoubtedly more “live and local” than ten great songs in a row can be.

The little factoids about the bands that play are interesting and decent, too. The only concern I’d have is that they’d need to be rotated quite sparingly to have the desired effect.

Sure, some phrases don’t sound quite right if you listen really carefully. But then, some phrases spoken by human beings don’t sound quite right either. (“This podcast is supported by advertising outside the UK”, which international audiences hear in front of most BBC podcasts, has the emphasis in the wrong place, skipping past the word ‘advertising’ and seemingly stressing the word ‘side’).

But, for an overnight shift, or a weekend service, or even to replace the non-stop music sweeps during the day, this is actually quite a passable listen, much though it pains me to say so. It’s worth a listen.

If you like AI-powered radio DJs, you’re going to get quite a few of them in the next six months, I suspect.

Spotify launched its own – which a) doesn’t sound like AI at all (and I suspect it’s been entirely faked for the press release’s video), and b) seems to contain nothing of any use, unlike RadioGPT. It’s a discombobulated voice telling you that “you listened to quite a lot of songs from the 1970s recently so here is another one” or “here’s a new release from a music type you love” without actually telling you the new release. Of all of the finesse of RadioGPT, this comes very lacking.

James Cridland

One Comment

  1. The industry should recognize what a watershed moment this is, a singularity even. Ignore it at your own peril. Naysayers teem in this business. Closed minds are the reason radio is in survival mode. Maybe with AI, we can find a way to turn it around.

Comments are closed.