How to change a few spoken words in a video
Upload the video to Vizard Agent and say which words should change and what they should become. Vizard Agent finds the phrase in the word-level transcript, clones the speaker's voice for the new wording, lip-syncs just those seconds, and cuts on a silent point in the waveform so the repair does not announce itself.
What do you type into Vizard Agent?
Quote both phrases. Vizard Agent finds the moment from the transcript rather than needing a timestamp from you, and the exact replacement wording is what lets it generate speech that fits the gap. Write dates and numbers the way they should be spoken, because "the seventeenth" and "August seventeenth" are different lengths of speech.
Prompt
What does Vizard Agent actually do?
One wrong date normally means filming again. Replacing a second and a half of speech instead means the voice has to match, the mouth has to match and the join has to be inaudible, so much of the work is verification. In the session behind this article it:
- Transcribed the video to word-level timings to find the exact phrase, analysed the speech around it for pauses and stutters, and extracted a clean audio sample as the voice reference.
- Cloned the speaker's voice for the new phrase with two different engines, then compared the clips for length and naturalness.
- Found the on-screen text box, cropped it to confirm the contents, and scanned the video at 0.1-second intervals to find exactly when the text appeared and disappeared.
- Lip-synced only the affected seconds to the new audio and analysed the result.
- Measured the audio energy around the join and cut on the sub-millisecond low point rather than a round number, so the join didn't click.
- Delivered a 744x1280, 60fps file of 17.26 seconds, the same length as the original, with the on-screen text rebuilt in a matching font.
When does this not work well?
The technical limits are real, and they are the smaller half. Changing someone's recorded words changes the record of what they said, and Vizard Agent can't know whether that person agreed, so it has to be your own video or you need permission. A much longer replacement won't fit, noisy originals clone poorly, and a close-up of a mouth shows any imperfection first.
Common questions
Will it sound like me? Vizard Agent clones from your own audio in the video, so it matches closely when the original recording is clean.
Does the whole video get regenerated? No. Vizard Agent replaces only the affected span and leaves the rest of your recording untouched.
What if the person is not facing the camera? Then no lip-sync is needed, and Vizard Agent replaces the audio alone, which is simpler and faster.