1 min read

I cloned my voice

I cloned my voice with Overdub, the text to speech software created by Descript, from the AI research division of Lyrebird AI.

You can do this too.

All you need is ten minutes and a microphone.

Overdub has created a base script to pull from to recreate your voice, so you won’t be lost for words but you will certainly be left speechless.

Why would anyone want to deepfake their own voice and how does the tech behind it actually work?

Let’s say you stumble over a word in an otherwise perfect recording, or maybe you have the bad habit of saying ‘like’ too much.

Overdub eliminates superfluous words to make you sound more smooth. It removes filler words to make you sound more professional.

But that’s not it, Overdub has also figured out how to add new words without having to record anything new and this is where things get really interesting.

Introducing text-informed speech inpainting, the technology behind generating missing audio as naturally as possible.

The challenges are that the audio needs to sound realistic, the sentences need to make sense and the digital quality must not suffer.

Descript wrote an in-depth, technical post explaining the imputations, which you can check it out here.