AI voice-over
Typing a line or turning your own read into another voice, choosing the voice and how it is said, and the take landing straight in the cut, made to fill the gap between your marks, with every take kept so you can switch between them.
Audio ▸ Voice-Over… turns a line you type into a spoken take, and puts it straight into your cut. The voices are made by ElevenLabs.
Making a take
1. Type what the voice should say. 2. Choose a voice. Hear This Voice plays a sample of it. 3. Press Make a Take. The button says how many characters it will use before you press it.
The take lands on the sound track shown under the button, at your in mark, or at the playhead if you haven't marked one. It goes in the AI voice-over bin, labelled as generated, with the voice, the words and the date.
How it is said
Expressive reads tags in square brackets. Tap a tag to put it in at the cursor: [bored], [excited], [whispers], [laughs], [sighs], [pause] and more. Put a tag before the words it should change. For example, [whispers] don't tell anyone is whispered.
Steady gives an even, dependable read, and speaks tags as words, so leave them out.
Under Delivery:
| Setting | What it changes |
|---|---|
| Stability | Lower is more varied and emotional; higher is steadier. |
| Similarity | How close to the voice as it was made. |
| Style | How much of the voice's own manner is exaggerated. |
| Speed | How fast it speaks, from 0.7 to 1.2. |
Fill In to Out
Mark an in and an out on the cut first, and tick Fill In to Out. The take is made to run exactly that long, by asking for the speaking speed that lands it. Before you press the button, the sheet shows how long the gap is and how long the line will roughly run.
Fitting can take a second attempt at the new speed, so the button says the most it can use. A voice only sounds natural between 0.7 and 1.2 times its speed. If the line can't fit inside that, the sheet says how many seconds over or under it is, so you can change the words.
Takes
Press Make a Take again to try another read, with different words, a different tag or a different voice. The new take replaces the one in the cut, as Take 2 of the same clip, and the clip keeps every take.
To switch, right-click the clip on the timeline: under Takes, the one playing is ticked, and choosing another plays it everywhere the clip is used. The sheet's list of takes does the same with Use This Take.
A take is always a new recording placed in your cut. B/side never changes words inside a recording you already have.
From your read
Record the read yourself first, as a guide: Audio ▸ Voice-Over… ▸ Record your own voice. When it is kept, press Turn It into an AI Voice…, or later choose Audio ▸ Voice-Over… ▸ Your read, in an AI voice.
Choose the read and a voice, and press Make an AI Take. The AI voice says your words with your timing and your emphasis, and the take goes into the cut in your read's place, as its next take. Your read stays as the first take: Use Your Read in the window, or the clip's right-click Takes, puts it back.
The take is used as the voice comes back. If it runs a little longer or shorter than your read, the window says so; trim it in the cut, or make another take.
Only reads you recorded yourself in the Voice-Over window can be turned into an AI voice, never an interview or somebody else's clip. A minute of read counts as 1,000 characters of your allowance.
How much you can make
Each paid plan includes characters of speech a month; Free includes 3,000 once. What is left shows at the bottom of the sheet. About 1,000 characters is a minute of speech.
Before each take, the sheet says what it costs: 420 characters: from your allowance, or, past it, the characters and the amount. Making the same line again with the same voice and settings gives you the take you already made, at no cost.
Posting
YouTube, TikTok and Instagram ask you to say when a voice in a video is made by AI. Check their rules when you post.