Is AI ready to read the news and weather on your radio station?
New AI services are appearing online ready to make it easier to run up-to-date information and alerts during live and automated programming.
Imagine AI being able to read a local / national mixed news bulletin, or weather forecast, or traffic alert, and automatically turn that text into an audio file ready for your playout software to squirt it into your log.
Imagine all of this happening at 3 in the morning whilst you sleep. Or whilst you’re busy on the other dozen projects you’re working on.
Of course – humans have been doing this for decades, but if you want AI to do it, imagine no more.
One such service to do exactly this has been developed in the Netherlands by Stefan de Groot and is now available to do all of the above and more for you. It’s called RadioNewsAI.com and is up and running after months of development.
However, having tested this service for the last couple of weeks, I’m not quite ready to deploy it on a permanent basis to any of my radio services.
It’s not because of the technology. The ability to add in text from sources such as your website’s RSS feed, or the local weather forecast, along with intro text, middle of bulletin text and an outro, comes as standard.
The ability to set the length of those gaps, add in news intros and beds, specify pronunciations, and add multiple sources to a single bulletin are most welcome.
And through pure magic, these bulletins can auto generate to your own schedule and import themselves into your log whenever you like.
Did I mention is re-writes the story rather than copy an article word for word?
But for me, it’s the voice that’s the current issue, particularly the British.
I think the American voices are almost there in terms of tone and intonation, and I’m using one to create hourly news bulletins for JingleMad Radio, which updates whenever there’s new jingle news on JingleMad.com. Will it remain? I’m undecided.
This entire tech and ability to create an up-to-date mixed local and national bulletin each hour by the power of AI would be perfect for a new project I’m working on – but I feel the voices need to be more human too.
Trusting AI to get the facts right, not inadvertently slander someone, and adhere to Ofcom’s broadcasting code, is quite a big risk for news. Easier with travel and weather, mind.
I’m more likely to use a service that can take a written bulletin and voice it word for word, so I know what will be coming out of the speaker at least. Maybe this can be added as an extra?
Here in the UK, Switch Radio’s Dean Kavanagh has developed his own system which is working for pulling in Met Office data Switch Radio’s local weather bulletins, and even powering the Weather 27/4 Radio service on DAB. This uses the openAI ChatGPT 3.5 service.
Both services are fun to explore, and one day I believe will be delivering hundreds of bulletins to radio stations around the world every hour.
In time, the voices will improve, the tech will adapt, and no doubt the ability to use your own choice of voice will appear in the near future, but for now, I’m keeping a very close eye on it, with the hope of using it for real one day.
Have a listen to some examples on the homepage at RadioNewsAI.com or click below to the news bulletin we generated below using the RadioToday RSS feed.
Roy Martin is CEO of RadioToday and all-round professional geek.