Reprinted from TidBITS by permission; reuse governed by Creative Commons license BY-NC-ND 3.0. TidBITS has offered years of thoughtful commentary on Apple and Internet topics. For free email subscriptions and access to the entire TidBITS archive, visit http://www.tidbits.com/ Text-to-Podcast Service Listen Later Sponsoring TidBITS Adam Engst We'd like to welcome as our latest sponsor [1]Listen Later, an Internet service that converts the text of any article on the Web into a custom podcast episode. You simply email the URL of the article you want to convert to Listen Later, and shortly after that, it shows up in your personal Listen Later feed in your favorite podcast app. Created by indie developer Yalim Gerger, who hails from Istanbul, Listen Later is a product of the modern age, relying on AI tools to process and narrate article text. Text cleanup is essential because many articles contain ads, image descriptions, disclaimers, and other addenda that get in the way of smoothly spoken audio. For instance, Listen Later is smart enough to avoid the comments at the end of TidBITS articles. And then there's the narration. We've come a long way from the speech synthesis days of Apple's Fred voice, and voice quality is shockingly good. Setup is simple'after you create an account, Listen Later's main URL displays your personal podcast URL, a field for the address it should associate with your feed, the email address to send to, and your available credit. Then it's just a matter of sending an email message'no Subject is necessary'to the address provided with one or more URLs in the body of the message. I first tested Listen Later on a 6700-word article by the author Patricia Lockwood, '[2]When I Met the Pope.' In retrospect, this may not have been the best sample text because Lockwood's prose flits descriptively from branch to glorious branch: At the reception ' no, excuse me, the 'vin d'honneur in the Lapidary Gallery' ' they are serving wine from Sting's vineyard. 'What did you say?' I yell at Hope, who is flying on a single glass, and she points to a brochure from which Sting stares out with a look of intense fermentation, Trudie's arms wrapped around his chest from behind, both of them wearing the honey of great good health and presumably fresh from a seven-hour stomping. I guess the idea is that you meet the pope and then get a mouthful of Sting's grapes. I nod to the brochure and Hope nods back conspiratorially, slipping it into her bag. That is how it is done; we are doing it. The Vermentino, yes, is called 'Message in a Bottle'. So is the Sangiovese. I had to go back and read chunks of the original piece to realize that Listen Later's AI narration had gotten everything right. Or almost right. Across a 36-minute audio version of the article, I noticed a couple of errors, mostly homographs like 'live' that have multiple pronunciations, although the narration also cut the word 'Sangiovese' short, eliminating its final vowel sound. I also thought it biffed 'Jesus,' but when I checked the text, I discovered Lockwood intentionally wrote the word as 'Jaysus.' In subsequent tests, Listen Later's phrasing continued to impress but occasionally stumbled on abbreviations and numbers, notably on prices like $2599, possibly because of our house style of omitting commas in four-digit prices for brevity. I don't know Patricia Lockwood, but the Listen Later voice on that first listen could have been hers, for all I knew. It wasn't, I soon discovered, but when I queried Gerger about it, he quickly added a setting that allows you to choose from six high-quality voices. My other early criticism of the audio quality was that the voice spoke just a hair faster than I would have liked. Initially, Gerger didn't think he could do anything about that, but a day later, he wrote back, saying that he had found and added a voice speed setting. In the meantime, I had discovered the speed problem was my fault. I listen to podcasts in Marco Arment's [3]Overcast, which has a Smart Speed option that shortens silences. It usually works well with people, who pause to breathe, but the AI narration was so smooth that removing silences made it sound slightly breathless. As you'll note in the screenshot above, Listen Later can even translate text into whatever language you prefer, although that costs extra. That's a good segue to talk about pricing. Listen Later has to pay for the API calls to process and narrate text, along with additional translation. Those fees directly relate to how much text you ask Listen Later to process. Your average article will probably cost somewhere between $0.30 and $1.50, and [4]Listen Later's pricing page provides a cost estimator so you can see how much any article will likely cost. Listen Later relies on a pay-as-you-go model rather than a subscription. Every new account gets $2 in free credits to trial the service, and after that, you can top up your account whenever you want. Or, if you don't want to think about it, you can set Listen Later to auto-refill by charging your credit card $5 whenever your balance falls below $0.10. This model seems fair and transparent, and those who decry subscriptions should welcome it. One thing to remember is that Listen Later, as befitting its name, isn't instant. The article's length plays a role in how long conversion takes, as does the load on the OpenAI servers, which is higher during US business hours. In the case of the lengthy Patricia Lockwood article, Gerger told me that it took about 20 minutes for the conversion to complete, and there were instances where his application waited up to 5 minutes for a response from the OpenAI servers. But this shouldn't generally be a problem'the entire point of Listen Later is that you're queuing an article up for listening to in the car on the way home, for instance, or during your next workout. Listen Later does include the URL to the original article in the podcast episode description for easy reference. I discovered that you can even convert PDFs to audio'I tested that with the [5]bonus chapter that Andy Weir recently wrote for [6]The Martian. Plus, Gerger just mentioned that he has added a feature to Listen Later that enables it to convert the text in images in JPEG, PNG, and WebP formats to audio. I was initially quizzical'what could be the use case? It turns out that certain communities commonly post screenshots of PDFs on X/Twitter. So, if that's of interest, just send Listen Later a URL to an image to convert it, just like a standard article. It's amusing that I'm writing about Listen Later, which converts text into podcasts, in the same week that Apple introduced a feature that converts podcasts into text. There are plenty of issues with generative AI, but some of the capabilities that it enables are magical. If you enjoy listening to podcasts and lack time to read everything you'd like, I encourage you to [7]try Listen Later with the $2 free credit every new account gets. References Visible links 1. https://www.listenlater.net/?referrer=tidbits 2. https://www.lrb.co.uk/the-paper/v45/n23/patricia-lockwood/diary 3. https://overcast.fm/ 4. https://www.listenlater.net/pricing 5. https://x.com/andyweirauthor/status/1756690693727154586?s=20 6. https://www.amazon.com/dp/0593357132/?tag=tidbitselectro00 7. https://www.listenlater.net/?referrer=tidbits Hidden links: 8. https://tidbits.com/wp/../uploads/2024/03/Listen-Later-splash.png 9. https://tidbits.com/wp/../uploads/2024/03/Listen-Later-main-screen.png 10. https://tidbits.com/wp/../uploads/2024/03/Listen-Later-settings.png 11. https://tidbits.com/wp/../uploads/2024/03/Listen-Later-in-Overcast-scaled.jpg .