Can AI help podcasts?

Last modified date

You can’t avoid conversations about Artificial Intelligence right now. For some, AI is the future that will streamline our lives and for others it’s the end of work as we know it. As it develops there can be few industries that won’t be touched by it, so it’s not a surprise that some are calling for the genie to be shoved back into the bottle. In the audio world, AI appears in audio editors that can transcribe and replace audio, construct artificial voices, compose music, and generate radio content. There are even tools to help write podcast show notes so you don’t have to.

This is either exiciting or terrying depending on how you view it, However, it’s also pretty focussed on producers, or at least those who would like to the expensive parts a little (or a lot) cheaper. None of this talks much about listeners, which where (if anywhere) AI might actually help us out.

As we know, when it comes to podcasting it feels like we’re riding a never ending wave of great content. If you ever chat to anyone who listens to podcasts and you’ll often find that sweet spot of a Venn diagram where you listen to the same stuff is pretty small. Dozens, if not hundreds, of new shows every week pop up and how do we know where to start? We might use the charts but these are basically indexes of attention and anyone who can throw cash at promotion or persuade the traditional media to give them coverage and they’re halfway there. I recently read an article that suggested the solution to finding your next favourite show was… newsletters. There’s logic here. Someone else trawls through a tonne of new shows, filters out the crap, and then gives you enough of a sense of each one to let you decide which you’ll search for and add to your app. To make best use of that we need to read a lot, filter our personal bias and for those newsletters to know that shows exisit in the first place and how do they know? I’m not dismissing the excellent work they do; columnists like Miranda Swayer have told me about lots of shows I’ve loved. The apps do some of this heavy lifting too but if a lot of this is based ‘people who like’ or genre based data then does that really cut through to what we like? Is this why we had a wave of new start-ups all promising to be the Netlfix of Podcasting, whilst totally missing the point about what either of them are and how they work.

If I think about what I listen to, it’s not really fixed by genre and I suspect a lot of other people are the same. What if, rather than guessing what we’d like to listen something else did all the hard work? Of course, people can do this and maybe that’s something we’ll all end up paying for but I wonder if AI could do a better job? If you look at the AI transcribers they can easily work out who’s talking and transcribe their, so it can’t be a big jump to work how many voices there are, how long they speak for, and whether they talk about football, Rishi Sunk, or the Godfather. This could mean it could work out whether the podcast it’s checking out is long-form interview, a panel discussion about league two football, or a complex layered documentary about motor racing. Can it look at the shows I like and not only spot some common like topics but also that I like crafted audio, decent sound, and nothing that’s going to drag on past an hour. Can it listen to a series and work out it’s usually 2 voices who talk to another voice, with some laughing? Maybe it can become super scary and start to recognise famous voices and tell us when they’re doing an interview? I generalise, of course, but the ability to peak inside what a podcast sounds like is surely going to throw up some better suggestions beyond listening the shows that are trending or are listened to by people who the platform thinks might be a bit like me?

(By the way, if anyone reads this and thinks it’s a totally brilliant million-dollar idea please do cut me in. if it’s already been done, send me a link)

Photo by Google DeepMind on Unsplash