How abcVoiceover works

1. 1 Create a project

   A project is one production: a podcast season, a game's dialogue, a course. It holds the scripts, the cast of voices that read them, the people you share it with and a discussion thread. Give it a name, a description if you like, and a colour so it's easy to spot in your list.

2. 2 Add your cast of voices

   Each voice in the cast is cloned from one recording. Press Record in browser and read for 5 to 15 seconds, or upload a WAV file, and name the voice after the part it plays: Narrator, Villain, Coach. The clip is all the model needs. Nothing is trained, so the voice is ready immediately, and you can replace its clip later.

3. 3 Write or paste a script

   A script is a list of lines, each with a voice. The script has a main voice, and any line can pick another from the cast. When you paste a whole script, start a line with a voice's name and a colon, like "Villain: You'll never catch me!", and that line is given to that voice. Lines can be up to 2,000 characters.

4. 4 Set the style for each voice

   Pick a style for each voice: Balanced, Expressive, Calm, or Fast draft for quick English takes. Choose one of 23 languages, and open Fine-tune to adjust emotion intensity, how closely the take follows the clip's pacing, and expressiveness. Settings are remembered per voice.

5. 5 Generate takes

   Press Generate on a line, or generate every new line in the script at once. Requests join a queue and are made in the background, in order, so you can keep writing or leave the page. Every take is checked by speech recognition, and a piece that skips words or runs on past the line is made again automatically.

6. 6 Compare and keep the best

   When a new take arrives, play it against the current one and keep whichever is better, or generate again. Saving a line locks in its text, its voice and the clip that voice was cloned from, so later changes to the cast never change finished lines.

7. 7 Share, review and download

   Share the project by account ID so collaborators can view it or edit it, and talk it over in the comments. When you're done, download the project as a zip: each voice's clip, and for every script its text, a CSV of who reads each line, and every kept take as a numbered WAV file in script order.

What makes it different from a text-to-speech box

Most text-to-speech tools take one block of text and give back one file. That's fine for a single narration, but it falls apart when a script has several characters, dozens of lines and a few rounds of rewrites. abcVoiceover keeps the structure of the script: who says each line, in what order, which take was kept and what it was made from. Changing one line means re-generating one line, not the whole file, and the export comes out already split and numbered for your editor or engine.

What you need

• A Google or Microsoft account to sign in with, and an invite code the first time.
• A recording of each voice: 5 to 15 seconds of one person speaking clearly, recorded in the browser or uploaded as WAV. See how to record a voice sample.
• Permission to use every voice you clone, if it isn't your own. See using AI voices responsibly.
• A script. Our guide to writing a script for AI voiceover covers what reads well.

The technology

Voices are made by Chatterbox, an open-source text-to-speech model with zero-shot voice cloning, running on a server the abcVoiceover team operates, with Whisper speech recognition checking every take. Your recordings and scripts aren't sent to an outside AI company and aren't used to train models. Questions about any of it are probably answered in the FAQ.