Change words without re-recording
Speech editing is coming soon. Highlight the words that are wrong, type what should have been said, and the recording says it in the speaker's own voice, matched to the room it was recorded in. Audio, or the audio track of a video.
Replace a word by typing it
A name said wrong. A price that changed after the take. A sentence that came out backwards. Today the fix is recording the line again and hoping it matches.
Instead: select the words, type what should have been said, and the recording comes back with them in it. Same voice, same pacing, same room.
On video, the picture is untouched. That suits voiceover, narration, and any shot where the mouth is not the subject.
Why edits usually give themselves away
We went through every way a wrong word gets fixed today, and what each one costs.
Tools that regenerate a word already exist. What they get wrong is not the voice. It is everything around it.
A cloned voice is clean. Your recording is not. Drop a studio-clean word into a kitchen recording and the ear catches it before you can say why.
We match the edit to the recording instead: the room, the distance, the noise underneath, the join on both sides. If you can hear where the edit is, it is not finished.
What you will be able to do
- Replace a word or a phrase
- Highlight it, type the new words, listen back.
- Fix a name or a number
- The most common edit, and the one a reshoot is most often made for.
- Take words out
- Remove a phrase and the gap closes without a click or a jump.
- Keep the recording you have
- Phone, room, noise and all. The edit matches the take, not a studio.
What you can use today
Two free tools, in the browser, no account. Each is a piece of the same engine.
Isolate a Speaker keeps one voice and removes everyone else. Enhance a Recording takes a rough recording of one person and returns it clean. What they learn from real recordings is what speech editing is built on.
Questions
- Is speech editing available now?
- No, it is coming soon. Join the waitlist and we will write when there is a recording you can try it on. Two pieces of the same engine are already free to use on this site: speaker isolation and speech enhancement.
- Can AI change a word in a recording?
- Yes. A model learns the speaker's voice from the recording, then says the new words in it. The voice is the easy part. The hard part is making those words sound recorded in the same room, on the same microphone, in the same breath as the words around them.
- Will the edit sound like the original speaker?
- That is the bar: the same voice, the same room, no audible join on either side. Where a recording does not clear it, the result will say so rather than hand you an edit you can hear.
- Does it work on video?
- On the audio track. The picture is untouched, so it suits voiceover, narration, podcasts published as video, and any shot where the mouth is not in close-up. A talking-head close-up will show the change on the lips.
- Does it work on rough recordings?
- That is the point of it. Phone recordings, echoey rooms and background noise are what people actually have, and an edit that only holds on a studio take is not much use. The edit is matched to the recording as it is, noise included.
- How much can I change?
- Short edits inside a sentence: a word, a name, a number, a phrase, or one sentence. Rewriting whole paragraphs in someone's voice is a different product, and not this one.
- How is this different from regenerating a word in a text-based editor?
- Text-based editors already regenerate a word from the transcript. What they hand back is a clean word in a recording that is not clean, so the edit becomes the tidiest moment in the take and the ear finds it. Here the new words carry the room and the noise around them.
- Whose recordings can be edited?
- Your own, and recordings you have the right to edit. It is built for fixing a take, not for putting words in someone's mouth. The speaker's consent is the rule, the same as for the voice cloning it depends on.
Hear it first
Speech editing is coming soon. Join the waitlist and we'll write when there's a recording you can try it on.