The What

We have launched an exciting new feature: AI-narrated stories. You’ll now notice a “Listen to this story” player just below the first paragraph of newly reported stories published after November 21. Press play, and you’ll hear a full narration of the story delivered in an AI-generated voice powered by OpenAI’s text-to-speech technology.

The Why

Earlier this year we launched Long Reads, an audio version of our indepth features. These episodes are narrated by our staff, produced with music and sound design, embedded on our features and distributed to third party podcast platforms like Spotify and Apple Podcasts. We created this format for readers who like to listen or consume content on the go and who might not have 25 minutes to sit and read, but can listen while commuting, cooking, or watching camels racing. “Cool,” we thought, “here’s a richly-produced, audio version for you to try.”

To our delight, the level of engagement with this format has been really high, with a lot of readers choosing to listen. This experience has demonstrated that audio storytelling improves the accessibility of our journalism and aligns with how a significant portion of our audience consumes media. 

However, as a small, budget-conscious nonprofit, producing a podcast-style narration for every story we publish really isn’t feasible. That’s where text-to-speech technology comes in. By automating narrations, we can bring this accessibility to our readers across a much broader range of content on our website.

The How

The world of text-to-speech technology has advanced rapidly in recent years, sometimes to a startling degree (listen for yourself and you’ll know what we mean!). In 2022, when we first explored adding narrated audio to our stories, the technology simply wasn’t ready. The available voices sounded robotic and clunky or the models made errors that disrupted the listening experience.

Fast forward to 2024, and the landscape has changed dramatically. After testing several leading text-to-speech services including Google’s Text-to-Speech, Amazon Polly, and OpenAI with our global team, we found OpenAI’s Shimmer voice to be the best fit for our needs.

It’s not perfect. You might notice the occasional mispronunciation, and the tone can sometimes feel mismatched for certain story contexts. However, overall, we think the listening experience is smooth and engaging. Give it a try and see for yourself.
One of the limitations we’re aware of with using OpenAI’s service is that the voices currently available are only in U.S. and English accents, and those voices are optimized for the English language. We’re keeping a close eye on future services and improvements so we can make our narrations more representative of our global audience and the regions we report on.

Feedback

We’re excited to launch this new feature, and your feedback is invaluable. Whether you’ve had a great experience with the narrations, noticed any issues, or have suggestions for better voice options, we want to hear from you! Email us at hello@restofworld.org with your thoughts.

Next

We plan to release this feature as an open-source WordPress plugin, making it accessible to other media organizations or developers who want to use AI narrations on their sites. If you’re a WordPress user or developer, stay tuned for updates in the new year via our newsletter. 

We hope you enjoy this new way to experience our stories. Happy listening!