What does AI mean for podcasts?

AI will change podcasting for good

AI is rapidly changing the world. What implications does this have for producers and listeners of podcasts? Plus: I used several AI tools in developing, producing and distributing podcasts in recent months. Where does it add value??

[This article was originally published in Dutch, at MarketingTribune]

During a historical development, it's often difficult to determine if you're part of it until after the fact. But you would say that surely it could hardly be otherwise than that 2023 will be the year AI had its breakthrough. Especially in the creative industries, the amount of smart tools is growing pretty much by the day. And if this year becomes the pivotal year, where we distinguish between the time before and after the emergence of AI, what does the future look like for the podcasting and audio world?

New possibilities

Many professionals have discovered the many AI tools available that make life easier. Within Airborne, we took the test last weeks and investigated where and how AI adds value within the process of podcasting (see below).

In the process, this new technology will lead to new user applications. Text-to-speech has existed for some time, of course, but AI makes it more affordable and accessible to a wide audience. Moreover, the quality of artificial voices is improving rapidly and are also increasingly easy to direct. With Murf.ai, for example, you can make adjustments fairly easily to get pauses, intonation and pronunciation right.

Creating a captivating audio experience with AI is still a challenge

Converting written text to audio is especially useful for those not in a position or mood to read. 'Audio as a service,' which is quite functional. Creating a captivating audio experience that draws listeners into a compelling story is still a challenge.

But if you think a step further, the ease with which you convert text into audio offers many starting points. To create personalized audio, you no longer have to go into the studio. A podcast where you address a listener by their own name, for example, is then possible. But think also of a customized newsfeed with the weather forecast in your region and the latest news about your favorite sports club. Or a high-quality production, in which certain segments are personalized - something that has of course already been done many times in print and audio. With the effect that the whole story feels personal.

Human connection

As a test, we had AI almost entirely create a podcast (see below). "Doesn't this precisely show us that podcast creators will soon no longer be needed?" we were asked here and there.

Now, as humanity, we are damned bad at assessing the potential consequences that technological innovations bring (there are those who believe that things have only gone downhill since the agricultural revolution). Thus, any forecasts about the impact of AI on our field and the world should be treated with caution.

AI mainly provides tools that facilitate and improve work

But for now, AI mainly provides tools that facilitate and improve work. So it is not taking over the work. Will that ever happen? Perhaps. But there are very many factors that determine the quality of a podcast: from the script to the host's voice, to the sound design, to the music. It would have to be a pretty ingenious tool that combines all of that and forges it into a fitting whole.

For now, the vast majority of listeners and creators actually see AI as a tool, according to a survey by podcast company Acast. Listeners also indicate that they find real people indispensable. Not very surprising: for now, a podcast is seen as an intimate medium, where human connection is essential.

6 WAYS AI CAN BE USED IN PODCAST PRODUCTION

Over the past few weeks, we have been exploring the ways in which AI tools can add value to our podcast production at Airborne. Here are six ways in which we have experimented with AI.

1. Concept Development

On Jan. 27, the world's first AI advertising agency was launched: The Uncreative Agency. Based on one-line briefing, this "agency" generates á la minute a deck you should be able to step under arm to your client with. The briefing: devise a podcast format with which a financial services company can profile itself among job seekers. The speed with which a deck (with 6 slides) is delivered is impressive. Unfortunately, the same is not true of the ideas: too little thought and too obvious.

By now it is clear that The Uncreative Agency is an initiative of advertising agency DDB. And of course the agency is well aware that their AI tool is not perfect, but we as a creative industry should embrace and develop AI - as it can only help us develop better ideas.

2. Finding Topics

For ANWB (Dutch Touring Club), we produce a successful podcast called 'Verborgen verhalen' (Hidden stories): a successful podcast in which we search for local mysteries and old legends. Finding new topics is always a matter of a lot of research, so a good time to see if that process can be accelerated with the help of ChatGPT. At first, the query for unsolved mysteries produces mostly a number of well-known missing persons and murder cases. But with better instructions - a "prompt," as it's called in jargon - better cases appear. At least so it seems. A story about the Dutch prince Willem III and his missing safe, full of valuables, is one of the greatest mysteries in Dutch history, ChatGPT reports. Equally interesting is the story from the 1920s, about the disappeared diamonds of the Dutch Ice Club.?

But it turns out to be a bit naive to pin our hopes here: further online research gives no clues in either case. None of the stories are true. And that is the Achilles heel of ChatGPT's widely used test version: information is presented in a firm, unqualified manner, while factual substantiation is lacking. Or, as the leading MIT Technology Review describes it, 'AI-language models are notorious bullshitters.' The researchers warn that we must be very careful with the tool in times of fake news and conspiracy theories.

3. Generated 98% by AI

Not everything ChatGPT spits out is nonsense. When we ask about the history of podcasts and possible applications, a true story does roll out. As a test, we then had this voiced by an AI voice from Murf.ai: in about 15 minutes, we had AI generate a podcast episode, with a beating story and a voice that was well-tolerated. So when you need to list generic and easily verifiable information, ChatGPT can be quite helpful in that regard. And then you can offer that as 'audio as a service'.?

Listen to the podcast episode generated by AI here.

4. Music

A good, narrative podcast without music is .... well ... not much really. It defines the signature of a production and is crucial to creating the right atmosphere. Beatoven.ai lets you have AI compose tracks at lightning speed, based on a number of variables. We had several tracks created, including one with the instructions: pop, fast and happy. The result is a track that you can easily model. For example, by switching different instruments on and off. The track itself is best described as 'stock music in its purest form'. And thus not nearly as useful. What is interesting is that you get a statement saying that the music is royalty-free.

5. Online Advertising

From Airborne we also develop campaigns to promote podcasts. And this requires online advertising assets: copy with a limited number of characters, of which several variants are needed, in order to test which ones convert the best. ChatGPT appears to be able to handle this as well. Based on a description and specifications for the number of characters available, several variants are generated. Here and there not quite the right tone, but that is just a matter of editing. This quickly saves an hour or two of text puzzling.??

6. Visuals

The cover art is the visual identity of a podcast and determines whether or not potential listeners will listen. On the AI front, you can easily have images generated based on the right prompt by Dall-E, from the same creators as ChatGPT. Another tool getting a lot of attention is Midjourney. This runs within the online platform Discord. Within various chat channels, you can share your instructions and then images are generated.

We are currently developing a true crime podcast, which revolves around a woman murdered in her apartment with a knife. We describe the environment and situation in as much detail as possible, in addition to the desired specifications of the image (such as: photo-realistic, 8k). The result is quite impressive. At least, when it comes to objects. Matching human figures with the right facial expression proves to be a lot trickier.