The Smart 7 goes AI
For the last week, we’ve been running a live experiment with our daily news podcast, The Smart 7.
If you’re not familiar with it, it’s a simple concept, the 7 biggest stories every morning at 7am, it’s always (about) 7 minutes in duration and we publish 7 days a week.
Jamie East was the original voice of the show, Ciara Revins voiced the Irish edition and since about July, she’s been hosting the UK one too.
But we began experimenting with AI voice cloning, partly out of curiosity more than anything and we trained an AI clone of Jamie’s voice. Basically, that means that we can type scripts into a window on the AI website and generate voice reads in Jamie’s natural tone.
Too Good not to try
It’s not perfect, but it’s scarily good at the same time and each time they update the software, it gets better. The big advantage for us is that we can instantly update stories as needed without having to wake up sleepy presenters!
We got to the point where we felt we had to try putting it live, so for the last week, Nessa McGann and I have been preparing the scripts and generating AI each morning, rather than producing a live voiceover. And, honestly, I don’t think anyone has really noticed the difference!
I’ll pop up links to two different episodes, one is with real Jamie and one with AI Jamie, have a listen and let me know in the comments what you think?
Smart 7 Example 1
Smart 7 Example 2
We’ve decided to keep the weekend episodes – That’s the Standout 7 on Saturday and the Sunday 7, our Science and Tech episode, voiced by a real live human, that is Ciara Revins.
How we make The Smart 7
Before we get to the pros and cons of AI Jamie, I’d like to just explain a little about how we make The Smart 7 each day. One of the first questions I get when I mention AI is, does AI write / curate the episodes?
The short answer is no, and I don’t think it would be able to for some time yet. Most of the data set in AI models like Chat GPT are quite out of date, so AI doesn’t really have a nose for news as yet. But it’s also quite a complicated set of editorial criteria to make The Smart 7 work.
The format is designed to be similar to a broadsheet newspaper in some ways, with the first two stores, (7 and 6), generally stories that are dominating the news cycle, so this last week, it was largely either Israel, or the Covid Inquiry.
5 and 4 tend to be either international stories, or inside page 2 or 3 of the paper, so we had Ukraine, the Donald Trump saga and King Charles in Kenya as stories last week.
Sport is at number 3, usually updated in the morning with audio, then 2 and 1 tend to be entertainment stories and maybe a trailer for a movie or a new show.
So, it’s quite a complex editorial mix, which Lucie Lewis pulls together through the day and we have two editorial meetings during the day to finalise. The scripts tend to get written in the evening and are then revised and polished or updated with new lines at 5am.
Also, each story has at least one piece of audio, so the script in total runs to about 3 or 4 minutes. That’s a large part of the reason that the decision to use AI was relatively straightforward, we wouldn’t be using it for any other format I don’t think.
Does this change anything?
So, what are the implications? Well, we’ve already seen a few rows kicking off over the perception that we’re replacing real life humans with robots, but it’s not really the case here.
Will AI / Voice cloning improve over time? Definitely. The first version of voice cloning I saw was a company called Lyrebird at one of the RadioDay’s events about three years ago and the development and sophistication of what we have today, makes that seem like stone age technology.
But where does it go next? I’m tempted to clone the voice of a radio presenter so that if someone calls in sick, or can’t make their shift, I could script and produce links to cover for them. I think it’s maybe still too early to do that and I don’t think many of my radio presenter friends would willing volunteer to be cloned as yet.
The Tech keeps coming…
Much like the US Actor’s strike, which is partly about AI and partly about residual payments, it’s a topic which is going to push its way to the forefront of the conversation as technology improves.
I can definitely hear the potential for an AI overnight presenter, but it’s not something I can hear working on a shift that has any form of content. I’m really curious too about whether or not people can detect the difference between an AI clone and the real person. There is definitely a clumsiness in phrasing or pacing in AI which I can hear now, but again, I think that will be corrected over time.
If you were running a small local station, how much of it would you consider using AI for?
This has been the 20th official edition of the RAudio Newsletter
Just to recap, each week I’ll be taking a look at big stories in radio, podcasting and audio.
Any feedback, questions or potential topics are welcome – you can get me on Linkedin https://www.linkedin.com/in/liamthompsonconsulting/ or on Twitter @Maxliam