RadioGPT, AI hosts and the future of human on-air talent

[kwthom]

Listening now---the male promo voice hasn't got the inflection down yet----the female voice is pretty solid. If I didn't know, I don't think I could tell.

I'd like to think there's some GM out there, listening to this demo right now, going "Yeah...that might be useable."

<...>What sort of return on investment would there be to have someone script out dialog, only to be repeated by a synthesized voice over radio? Text to speech doesn't translate inflection, tone, emphasis, or if a human smiles while they talk. Any of that would require much more specific coding for every sentence spoken by a synthetic personality. Something like that would be well out of the financial benefit for any radio station or group.

Teaching the 'computer algorithm' to be a radio announcer may not be the challenge some think it to be. Certainly, there will be some growing pains, and might not make compelling 24/7 radio, but...

Could that be 'passable' from 10p to 5a on radio stations in the bottom 200 markets?

 

[michael hagerty]

I'd like to think there's some GM out there, listening to this demo right now, going "Yeah...that might be useable."

Teaching the 'computer algorithm' to be a radio announcer may not be the challenge some think it to be. Certainly, there will be some growing pains, and might not make compelling 24/7 radio, but...

Could that be 'passable' from 10p to 5a on radio stations in the bottom 200 markets?

I hate to say it, but it's better than a lot of what I've heard below market 30, in any daypart. And it's only going to improve.

 

[michael hagerty]

...I mean, part of the problem is that we've let radio get to a point where what's on is not a lot different from this in the first place, but we're here, we've been here for a couple of decades, and apart from a few exceptions (Broadway Bill Lee), this isn't that big a leap.

 

[kwthom]

I hate to say it, but it's better than a lot of what I've heard below market 30, in any daypart. And it's only going to improve.

I'm thinking it's only going to take a few pro's to hear this and really understand that this *could* be the next step beyond VT.

Listening for the last ten minutes, it's not quite polished, but it's certainly analogous to a radio professional that's 'on' every time that mic is opened. The patter between the male and female voice is a little rushed at times (tight? Oh, heck yeah!) and maybe that works in Big City, OH, but maybe not so much in Oil Crank, WY.

 

[michael hagerty]

I'm thinking it's only going to take a few pro's to hear this and really understand that this *could* be the next step beyond VT.

Listening for the last ten minutes, it's not quite polished, but it's certainly analogous to a radio professional that's 'on' every time that mic is opened. The patter between the male and female voice is a little rushed at times (tight? Oh, heck yeah!) and maybe that works in Big City, OH, but maybe not so much in Oil Crank, WY.

I’m sure that’s adjustable. I went a little way into the “more information” pages. You can specify male talent, female talent or both.

 

[TheBigA]

What all this really should do is cause those who call themselves talent to re-evaluate what they do in context of value to employers. If all they do is back-sell songs, that's not enough. If all they do is read scripted liners, that's not enough. There are lots of things talent can do that is unique to being human. Anyon who does this should be able to identify those things and focus on them.

 

[michael hagerty]

What all this really should do is cause those who call themselves talent to re-evaluate what they do in context of value to employers. If all they do is back-sell songs, that's not enough. If all they do is read scripted liners, that's not enough. There are lots of things talent can do that is unique to being human. Anyon who does this should be able to identify those things and focus on them.

We do need to stipulate, though, that a lot of talent only does those very basic things because their PDs and GMs insist on it.

 

[TheBigA]

We do need to stipulate, though, that a lot of talent only does those very basic things because their PDs and GMs insist on it.

They don't stipulate things like tone or inflection or character. Those are left to the performer or interpreter. That's what I'm talking about. If you're just going to read someone else's script without providing personality, then we have a tool for that.

 

[Theater of My Mind]

I suspect the type of radio station owner/manager who would use this sort of automation is not someone who cares very much about talent, regardless of how much 'value' that talent offers to the employer. Their number one priority is inevitably about paying as little as possible for the content they broadcast.

 

[TheBigA]

I suspect the type of radio station owner/manager who would use this sort of automation is not someone who cares very much about talent, regardless of how much 'value' that talent offers to the employer. Their number one priority is inevitably about paying as little as possible for the content they broadcast.

I don't think this device is free to radio or without the need for programming. So there is a cost involved. Just as there is a cost in running syndication or even VT. The balancing act is in terms of return on investment. Paying a salary & benefits is a form of investment. On the talent side, the challenge is in pricing your services so they're competitive with other options without working for free. I think moving forward this will be the challenge for employees in a lot of businesses, because the cost of benefits is so high. So if all you're doing as talent is providing human continuity, that's probably not a full time job, and you can combine it with either other voice work or other types of work.

 

[wadio]

Regarding the future of AI radio, there are two sides to the equation:

• What AI is capable of
• What listeners will accept

There was a time when no one could be expected to watch a video shot with a handheld camera, or accept Zoom-quality audio over broadcast media. Today those things are completely acceptable.

As lower quality content is rolled out, consumer expectations are gradually lowered. That will likely happen with AI, sooner than we expect.

 

[kwthom]

I suspect the type of radio station owner/manager who would use this sort of automation is not someone who cares very much about talent, regardless of how much 'value' that talent offers to the employer. Their number one priority is inevitably about paying as little as possible for the content they broadcast.

How many of those stations already exist?

I don't know what costs are (costs to have someone live, costs to pay someone to VT a show, costs to have ChatGPT provide programming), but if they are 50% cheaper than paying someone to VT a shift (mid-day? evenings?) , then I think that radio station owner/manager might certainly consider the possibility.

Someone will be first, while the rest sit and watch to see what John Q Public thinks.

 

[michael hagerty]

How many of those stations already exist?

I don't know what costs are (costs to have someone live, costs to pay someone to VT a show, costs to have ChatGPT provide programming), but if they are 50% cheaper than paying someone to VT a shift (mid-day? evenings?) , then I think that radio station owner/manager might certainly consider the possibility.

Someone will be first, while the rest sit and watch to see what John Q Public thinks.

....and, because some of the comments sound like people haven't clicked the link and listened to the demo, do that.

 

[TheBigA]

Getting back to what I was saying about value for the employer, here are things the AI DJ can't do:

1) Personal appearances at local concerts, malls, or events.
2) Direct interaction with listeners via social media, text, or even phone.
3) Make sales calls with the AE, to create that personal connection between what's on the air and the client.
4) React to immediate emergencies in the community.

Those are just a few that came to mind. Obviously some GMs or PDs don't care about any of that. Just play another 10 in a row. But on the other hand, when you have 30 stations in a market competing for the same audience, you need to differentiate in some way.

 

[vchimpanzee]

That first one is done, or has been done, by the local oldies station which was America's Best Music before 2015. There is also local news and information on the morning show. As for number four, there was a brief interruption in the music for one of the worst fires in recent memory, which is surprising because that would seem to be a situation where they would go on the air with news for an extended period. We've heard many times about the two firefighters who died who were honored several times.

 

[gr8oldies]

It didn't sound all that bad, and I don't know that it's worse than a lot I hear on the air. It can do more than the voiceover on automation reels did decades ago. We shall see.

 

[michael hagerty]

It didn't sound all that bad, and I don't know that it's worse than a lot I hear on the air. It can do more than the voiceover on automation reels did decades ago. We shall see.

I actually went back and listened to a Drake-Chenault automation demo after hearing the AI demo. This could easily have done that, and as you say, more.

 

[Kelly A]

How many of those stations already exist?

I don't know what costs are (costs to have someone live, costs to pay someone to VT a show, costs to have ChatGPT provide programming), but if they are 50% cheaper than paying someone to VT a shift (mid-day? evenings?) , then I think that radio station owner/manager might certainly consider the possibility.

But none of that is the case. ChatGPT is a textual chatbot, not an announcer. It doesn't even have a voice. And synthetic voice is not AI.

 

[PTBoardOp94]

I'd like to think there's some GM out there, listening to this demo right now, going "Yeah...that might be useable."

Assuming the price is right, I'd think stations using syndicated formats would be good candidates.

If you're using Westwood One's 24/7 formats, you get maybe 60 or 70 seconds of jock content in an hour, and most of it is backselling records, with the occasional bit of trivia. Hard for them to do much more, given that they are neither live nor local, nor do they operate contests or take phoners.

Could a chatbot do that? Probably. If not today, in another year or two.

 

[gr8oldies]

Assuming the price is right, I'd think stations using syndicated formats would be good candidates.

If you're using Westwood One's 24/7 formats, you get maybe 60 or 70 seconds of jock content in an hour, and most of it is backselling records, with the occasional bit of trivia. Hard for them to do much more, given that they are neither live nor local, nor do they operate contests or take phoners.

Could a chatbot do that? Probably. If not today, in another year or two.

Westwood One will do Custom breaks, but this AI would know instantly (if programmed correctly) the local football score. "At halftime, State U. leads West Podunk State 21-0"