Ross ZeigerHi, I’m Ross. Former Marine Corps officer, business founder, and YouTube expert — here to make video simple.

Descript's AI Text to Speech Feature (aka Overdub)

2 Oct 2023 · 3 min

Watch on YouTube

2:30

A short tutorial/demo describing Descript’s Overdub text‑to‑speech feature: how to train an AI model on a user’s voice (recording or uploading speech), submit a voice ID for verification, and use Overdub inside Descript’s editor to replace highlighted text with synthesized audio. The creator notes training-time recommendations and a brief walkthrough of writing and replacing text with Overdub audio.

Questions this video answers

At least 10 minutes (they recommend 30 minutes of English-language speech).

About 15 minutes.

Press W to enter write mode.

Submit a voice ID (a short signed/verified recording asserting you are the speaker) which Descript checks against the training data.

This isn't me talking. This is an AI version of me. Descript's most popular feature is the Overdub feature, also known as the text to speech feature. You can train the AI on your voice so that it can replicate it by text like you just heard at the opening of this video. The way you do that is in your descript. You come to voices and you come to this page. I've already trained it, my voice. So if I open it, this is what it looks like. I read a essay for 15 minutes and that was enough to give it the voice. And it's not great. They recommend that you train it for 30 minutes or more and they even have a script that they recommend you read. Oh, and here it is on the side. They upload or record at least 10 minutes. We recommend 30 of English language speech in your voice. You can also paste media from existing Descript projects. So if you've already got a podcast or video where you are talking for an extended period of time, you can upload that and it can use it as your training data. And then you have to submit your voice id. So what that is is they ask you to read a short thing saying that I verify this is my voice and I'm not stealing anyone else's data because they don't want you to record some celebrity or your friend and use it in malicious ways. That is why they do that. And they will verify that those voices match the training data and the voice id. And then it takes some time to process and you'll get an email saying, congratulations, your Overdub voice is ready. And they have some tricks and tips and tricks here to make your voice as good as possible to use the Overdub voice. Here is the project that I'm recording what you're listening to me in right now. So what you do is you can press W and that enters write mode. And you can start writing something that you want Overdub to say in your voice, just like that. Exit writing mode. And what you would do is highlight it. Highlight the words that you want it to say. This menu pops up over the highlighted words. You hit replace. Oh, and it's already converted it to audio. Look at that. But if it hadn't done that, you would hit overdub and it would process it for a moment. It'll appear in gray while it's processing. And then you get your voice here and let's go ahead and listen to that and you can start writing something that you want Overdub to say in your voice. Just like that. Super easy to use. That is Descript's text to speech feature.