Descript's Overdub feature lets you make a deepfake of your own voice

The podcast editing service can also use AI to remove verbal ticks or filler words, significantly reducing post-production.

A man can be seen in front of a microphone. He is sitting in front of a desk.
AleksandarNakic/E+/Getty Images

Podcast editing service Descript has announced a new tier of its subscription offerings called Descript Pro that includes a new feature which, until now, has been in beta. Called Overdub, the feature allows you to create a virtual model of your own voice, that can then be used to automatically fix audio mistakes and hiccups on podcasts.

The audio editing software, which relies on artificial intelligence, is launching the new feature today. Overdub lets you replace text in the automatic transcripts of audio recordings it generates, and then uses the model of your voice to replace to the offending snippet of the recording with the replacement text you've entered.


How it works — The company describes its Overdub feature an opportunity to create an "ultra-realistic clone" of your very own voice. It's essentially a sanctioned, audio deepfake of your own voice. It's very handy for correcting mistakes and removing umms, ahhs, and other verbal ticks. But we can't help but wonder what a nefarious person might be able to do with a model of someone influential's voice (not that you aren't influential, that's not what we're saying).

All it takes, according to Descript, is a solid mic and 10 minutes of your voice reciting a Descript script to train Overdub. Ideally, however, the company suggests that you give Overdub 30 to 90 minutes of your voice to help it create an even more accurate model. As it stands, you can't create an Overdub clone of someone else's voice... at least not easily. Presumably, you'd need to trick them into letting your record the training script and then feed it into the app yourself.

Bonus round — An intriguing chunk of linguistics' literature focuses on filler words. You know, um, those words that, uh, some people use in their, err, speech patterns. Scholars suggest that these words have their own social utility and might be reflective of people who are cautious in public environments. In the professional world, including podcast-making, filler words are often discouraged. Descript helps eliminate them with its Filler Words Pro feature which it also offers as part of its subscription services. The tool allows you to pick up on and even remove filler words as shown below.


So, not only will artificial intelligence attempt to emulate your audio personality, but Descript could also help save you heaps of time in postproduction. On top of that, the company's subscription service has improved upon its transcription abilities, file export features, flexibility around publishing, and customization options.

Podcasts may not be quite as popular as they were before COVID-19 lockdowns, but Descript seems to have the potential to strongly improve the genre's game. It's also offering the sort of unique services that make us wonder whether we might not soon see it acquired by Spotify, which definitely believes there's plenty of growth to be hadand money to be made — from podcasts.