It is still difficult to qualify just how widely adopted AI voice use is across the radio industry.
We have come across quite a few implementations for services such as weather forecasts, particularly in smaller markets. On the other end of the spectrum is iHeartMedia’s recent “guaranteed human” pledge.
A study from 2024 found that about one in five radio listeners think they’ve heard AI-generated content on the air. It’d be fascinating to see results today.
That said, with it being three years since the first wave of radio adoption of synthetic voices began, would you be able to even tell the difference if you heard AI on the air?
A recent study conducted by the University of Mississippi’s Jordan Center for Journalism Advocacy and Innovation inspired Radio World contributor and friend Dan Slentz to send us six underwriting announcements he produced for 105.9 WDOG(LP) in New Philadelphia, Ohio.
Slentz serves as the station’s consultant. The rock-formatted community station airs four underwriting announcements per hour, from the hours of 6 a.m. through midnight. That could, if “The Rock Dog’s” inventory is filled, total a maximum of 72 announcements per day.
“Sometimes getting a different voice in the mix is difficult,” Slentz said.
[Related: “What Happens When AI Shows Up on Your Station’s Social Media?”]
So, to enhance the station’s narrators, he has experimented with low-cost or free AI voice resources. Most recently, Slentz has used the free AI voice generator offered by QuillBot, which advertises speech that sounds “natural, expressive and human-like.”
Below are six different MP3s of WDOG underwriting announcements.
Three are recorded by humans. The other three are from AI.
Listen to them first, and then we’ll reveal their sources.
Human or AI?
- Bear Carpet One Floor & Home in Sugarcreek, Ohio
- Berner Trucking in Dover, Ohio
- Better Banner Printing in New Philadelphia, Ohio
- Dover Flea Market in Dover, Ohio
- Dundee Marketplace in Dover, Ohio
- Michael’s Restaurant in New Philadelphia, Ohio
Radio World’s performance
Before we reveal the answers — as Professor Slentz, also an instructor and director of John Carroll University’s Blue Streak Media, informed me — yours truly would have failed this test if it were being graded on an academic scale.
I had guessed that Berner Trucking, Better Banner and Dover Flea Market were AI. Just two of my six guesses were correct!
Answers revealed
- Human: 3, 4 and 6
- AI: 1, 2 and 5
The Better Banner announcement, Slentz told us, is voiced by Steve Kelly.
Michael’s Restaurant is the voice of Dan’s wife, Gina.
Slentz himself is heard on the Dover Flea Market announcement.
The other three announcements? They are generated by AI.
“It’s pretty amazing how realistic AI can sound,” he told us.
Science backs the notion that we’re having a harder time distinguishing the difference between human and AI voices. A study conducted by Nadine Lavan, Mairi Irvine, Victor Rosi and Carolyn McGettigan that was published last September in the National Library of Medicine said that voice clones — similar to what WDOG uses — are labeled as human between approximately 58 and 70% of the time. AI voices, conversely, are labeled as human approximately 40% of the time.
The researchers concluded that “the widespread availability of human-like AI-generated voices as created by commercially available products has numerous implications for how humans evaluate the voices they hear around them.”
How did you do on the quiz? And have you deployed AI voice tools at your station? Let us know, as we’ll include it in a Radio World follow-up story.
[Read More Radio World Stories About Artificial Intelligence]
