Fix a finished performance by editing text — not re-recording.
Change the script, and MendWave regenerates only the altered span — in the original performer's voice, inside the original take. Every candidate is scored — word accuracy, speaker match, naturalness — and ranked before you hear it; no voice is stored or cloned.
Research preview — free, single-word fixes. No signup, no credit card. Output is not watermarked. No voice cloning — performer consent required.
Playing the original take. Switch to “After” to hear the v0 edit — one word regenerated, everything else the untouched original. Demo audio: LibriTTS (CC BY 4.0).
Each edit is transcribed back and checked word-for-word against your revised script.
A speaker model compares the edit to the surrounding take, so the voice stays the performer's.
Loudness-matched, room tone preserved, crossfaded at low-energy cut points.
Eight takes are generated and scored; the ranking and every score are shown, so a weak best-take is visible, not hidden.
Three steps. No booth, no recall.
Getting talent back in a booth is expensive — when it's possible at all. Cloning their voice is a line studios rightly won't cross. MendWave is the third option: regenerate only the region you edited.
Upload the take, edit the script
Drop in the finished audio and its script (or auto-transcribe), then click the word to fix and type its replacement. Whole-script diffing is next on the roadmap.
Only the edited span is regenerated
The altered region is re-synthesized in the performer's own voice and spliced back at natural pause points. Everything outside the edit is still the original take.
Scored — every candidate
Eight candidates are generated, each re-transcribed, speaker-matched, and scored for naturalness — you get the top pick plus the score table for every candidate.
Why it still sounds like the performance
The difference between clicking a word and hoping — and an edit you'd put your name on.
Verified, not vibes
Every candidate is re-transcribed, speaker-matched against the take itself, and scored for naturalness and pacing. The scores ship with the audio — you judge with the numbers in hand.
Script-first editing
Bring the script and edit against it — today one word at a time in the v0 preview; whole-session script-diff batch is the next milestone on the public roadmap.
No voice cloning
We verify each edit against the original performer with a speaker model — but build, store, and sell no voiceprint. Performer consent first; the master is never altered outside the splice.
Invisible splices
Cuts land on low-energy points, levels are loudness-matched, and the room tone carries through the edit — so the seam doesn't announce itself.
An edit report, every time
Each job ships with a line-by-line receipt: what changed, the candidates considered, and the verification scores. Provenance you can forward to talent, an agency, or a rights holder.
The master is sacred
Audio outside the spliced region is bit-for-bit your original file. No re-encoding sweep, no "enhancement" you didn't ask for.
Made for the takes you can't get back
Wherever the performance is the product and the performer isn't in the building anymore.
ADR without the recall
A changed line, a legal note, a late script fix — after the actor has wrapped. Edit the dialogue inside the original performance instead of chasing a pickup across schedules and continents.
VO that keeps up with the build
Renamed characters, rebalanced quests, live-ops updates — thousands of recorded lines drift out of date the day the sessions wrap. Fix them in the recorded voice, patch by patch, without another session.
Fixes in every language
When the translation changes after the dub is in the can, the fix shouldn't need the studio again. Adjust the dubbed line in the dub actor's own voice — and keep the performance they gave.
Also at home in audio drama and audiobooks — where MendWave started.
One price. $15 per fix.
What would it cost you to redo this recording — the re-booking, the studio time, the blown deadline? That's the number we price under.
First fix free
The v0 research preview is free — full pipeline, real scoring. Preview output is not watermarked; watermarking lands with the paid launch. Proof it works on your audio before a card ever comes out.
$15 per fix
Prepaid, any quantity, same unit price. A fix is one edit job: change the script, get the verified result in the performer's own voice, with the edit report as your receipt.
No tiers, no subscriptions
No plans to compare, no volume ladders, no meters counting your minutes. Buy three fixes or three hundred — the price doesn't move.
Identity continuity — not voice cloning.
Tools that can speak in someone's voice deserve suspicion. MendWave is built so the answer to "where did my voice go?" is always: nowhere.
- ✓Performer consent is required before any edit is generated.
- ✓No voice profile is trained, stored, or sold — each edit is checked against the surrounding audio only, then the comparison is discarded.
- ✓Your audio is never used to train our models.
- ✓Every regenerated span is speaker-verified against the performer's own surrounding audio before it ships.
- ✓Flagged edits stop at review — nothing unverified reaches your master. The edit report is provenance you can hand to talent or an agency.
- ✓Demos and testing use public-domain LibriVox recordings.
Fair questions
Will listeners hear the edit?
Is this voice cloning?
Can it handle expressive, emotional delivery?
What do I upload?
What happens if an edit can't be verified?
Do you keep or train on my audio?
What does it cost?
Hear it fix your take
Open the editor, upload a short clip, type the change, and listen to the verified edit in the performer's voice — before you've finished your coffee.
Open the editor →First fix free · then $15 per fix, prepaid · performer consent required · output at 24 kHz.