AI voice production has been one of the most discussed topics at this year’s NAB Show, and radioinfo’s Wayne Stamm talked to Jeanette Kadow from Veritone about their synthetic voice product.
Veritone won product of the year at the NAB shjow in 2021 with it’s voice cloning product that can either clone a voice or create an exact replica.
radioinfo: So let’s take a look at Veritone’s AI Voice generation.
Jeanette: Yep. So we won product of the year at the NAB in 2021 with our synthetic voice product.
What we can do is we can either clone a voice and create an exact replica of a talent’s voice so that it cuts down on them having to do studio time in reads and promos and all that stuff. We can literally tell it what to say and it comes out speaking like Ryan Seacrest or whomever, or we also have a library of stock voices with, you know different male, female children, age ranges, different languages, different dialects of those languages or accents.
Australian accent is one of the versions here in English, and that gives stations who might only have 1 or 2 voice people the option to now have several voices on the radio or more voices of, you know, voicing the the advertisement. And it could also be in multiple languages.
radioinfo: One of the things that I heard discussed not long ago was then are the extraordinary lengths that the Veritone goes to to make sure that somebody’s voice isn’t actually stolen.
Jeanette: That’s right. So that so that’s a really good point. So you can go on the Internet today and create a voice in a minute, and there’s no ethical security around that at all.
We are we are an enterprise grade voice solution. We have inaudible watermarking on that voice, there’s permissioning on that voice. We consider it IP. We take that very, very seriously.
radioinfo: And you’re seeing it continue to expand, demand for it is growing?
Jeanette: Absolutely. So I think when it first came to the radio market, you know, the DJs were afraid that this was going to take their job until they started realising, wow, I really only have time to do these five, you know, clients voice overs, but now I can do 50 for across the entire country and I can actually get paid for it.
Now they’re monetising their own brand and their IP.
I’ve seen just a huge difference in the last six months of it, you know, it being more widely adopted.
radioinfo: I heard, a term that was used about how Veritone is able to generate the voice and get it so good. Was that Garbage in-Garbage out?
Jeanette: Yep. So garbage in. Garbage out. So again, if you go to kind of like one of these websites where you can create a voice in a minute, you’re going to get garbage.
It takes about three hours of training data to actually create a broadcast quality voice if it’s clean audio, and radio has clean audio. And then we just synthesise it and then it’s basically, you know, utilised at your discretion with permissioning and consent.
Veritone is first and foremost and AI company that has built an operating system for artificial intelligence that leverages what they regard as all the world’s best AI models into one operating system.
In their operating system they have real time transcription for radio broadcasts, text recognition, speaker separation and content analytics.
Veritone work with about 90% of the radio broadcasters here in the US, ingesting their live terrestrial feed and it gets indexed with real time transcription landing on their platform roughly about ten minutes after it airs.
Basically this means that anything said on the air is now searchable.
To hear more about what they can do with their analytics, and their application called Discovery, listen to the following from Jeanette.