Video EditingIntermediate

How to Fix Mistakes with Overdub AI Voice in Descript

Train an Overdub voice and patch a misspoken word by typing the correction, so you do not have to re-record.

9 minIntermediate

Overdub is Descript's voice cloning feature. After you train it on samples of your voice, you can fix a wrong word or add a short phrase by typing it, and Descript generates audio in your own voice. This is a lifesaver when re-recording is not practical. This guide covers training and a basic correction.

What you need

  • A Descript plan that includes Overdub
  • About 10 minutes of clean recordings of your own voice for training
  • Consent: you can only legally clone a voice you own or have permission to use
  • A project with a small spoken mistake to fix

Step 1: Create and train your Overdub voice

Go to Settings, then Voices or Overdub, and click Create a voice. Descript has you record a consent statement (proving it is your voice) and then read training script lines or upload existing clean audio. Aim for at least 10 minutes of clear speech. More training data gives a more convincing clone.

Descript - Overdub training
Create Overdub voice
------------------------------------
Step 1: Record consent statement
"I authorize the creation of..." [ Recording 0:14 ]
Step 2: Read training lines (0 / 50)
Step 3: Train model (locked)
Tip: a quiet room and your usual mic = best clone
Recording the consent statement before training begins.
Voice cloning needs consent
Only train Overdub on a voice you are allowed to use. Cloning someone else's voice without permission is against Descript's terms and can be illegal. The consent step exists for this reason.

Step 2: Find the word to fix in the transcript

Open your project and locate the misspoken word in the transcript. Say you announced the price as ninety dollars but it should be nineteen. Highlight the wrong word so Descript knows exactly what to replace.

Step 3: Replace it with Overdub

With the wrong word selected, type the correct text right over it. Descript detects an Overdub opportunity, or you can right click and choose Overdub. Pick your trained voice and generate. It synthesizes the new word in your voice and splices it into the audio.

Descript - Overdub replace
You
The plan starts at just [ninety] dollars a month.
Agent
Overdub: replace 'ninety' with -> nineteen | Voice: My Voice | Generate
The transcript with the wrong word selected and the Overdub popover open.
Overdub generation
Selected: 'ninety' -> typed: 'nineteen'
Overdub voice: My Voice
Generating audio for 'nineteen'...
Inserted at 00:42. Match confidence: high
$

Step 4: Listen and adjust pacing

Play the sentence back. If the generated word feels rushed or oddly emphasized, Descript lets you regenerate or nudge the timing. Sometimes adding a tiny pause before or after helps the patch blend with the surrounding real audio.

Result: a pricing error in a finished voiceover fixed in under a minute by retyping one word, with no need to set up the mic and re-record the whole take.

Watch related tutorials

Tags
#descript#overdub#ai-voice#voice-cloning