Radio Drops #5: The A.I. Jingle

A.I. Voices Aren’t Yet Human. Jingles are Close Enough.

We are all actors. Not just us media people, all humans are acting out their own lives. We behave differently and we talk differently depending on the situation. We never, ever talk ‘perfectly’. Yes, today’s audio technician geek can replicate a person’s voice well enough to fool 90% of people, but it still takes a lot of time to get it right. To do that in the long run, to have (say) a “live” DJ every day that regular listeners think is a human? I haven’t heard it yet. What happens when they learn they’ve been listening to an android? Are they going to feel duped? We’ll soon find out.

The problem with creating a voice that will pass as human is the breath. That’s how we can all tell the voice of a machine. An A.I. doesn’t breathe and if it don’t breathe, it ain’t alive! Meanwhile on the production side of things, the new tool is getting amazing. The radio manager in Oslo created this one evening in his home office, just messin’ around.

Watch the video on YouTube: Radio Drops #5: AI Jingle

Can You Still Tell It’s A.I.?

The day I record this, I sent a voice sample of myself to a pal of mine who’s going to run it through A.I. and see if it can create a brand new Steve Warren, one who can talk about things I don’t even want to talk about. We’ll see how that comes out.

If I were still in the voice business, making a living with my voice, I would be a little concerned about these A.I. voices. The advertising agencies, and yes, the mega broadcasters, are going to jump into this thing feet first, because hey, we can eliminate somebody’s salary with it.

But research is already showing that everybody can still tell the difference. Radio World had a headline on March 27th about a national survey, pretty big sample: nearly half of those surveyed had an outright distaste for A.I. voices. We’ll still be able to tell that’s not a real person talking to us, for a while.

In late March or early April 2024, CBS had a six- or seven-minute profile on a Sunday morning show of country singer Randy Travis. He lost his voice to a brain aneurysm, I think it was, some years ago and cannot sing anymore.

He got together with his longtime producer and his wife, and they took samples of the old Randy Travis singing, just the voice, and made a new record with it. It sounds dadgum close. People were coming to tears listening to it. But it took them a long time and a lot of work to replicate his voice.

However, one evening a few months ago, sitting in his home office, a radio manager in Oslo typed up a jingle description for his station and told A.I. to make a jingle. Here’s what it came out with, and this is pretty good.

If you didn’t know, this type of jingle is called a donut. It’s got a little musical interlude in the middle. I love the thing where the disc jockey can jump in and say, “You’re listening to the Beat.” It’s in Norway, it’s in Oslo, and it’s called the Beat.

Links to the A.I. Toys: ElevenLabs and Suno

In my recent posts I’ve had examples of A.I. created jingles and my voice replicant (Radio Drops #6). That was made at ElevenLabs.com. Here is a link to the site.

The jingle came from the ‘make a song about anything’ site Suno.com and here is a link.

I’ve not put any time into the voice generator myself, but I did toy around with Suno. I asked it to make me an a-capella jingle for “Magic 96” that I could drop over song intros. That’s all I wanted, just three or four voices singing those five syllables. “Magic Ninety Six”. I could not get Suno to do it for me; it kept writing me songs about Magic 96. I spent only about 15 minutes with it, however so maybe it is possible. If you get entranced with it and find the magic phrases and instructions to sing call-letters or station brand names, let me know.