How to match how someone speaks and not just how their voice sounds
Tell Vizard Agent to match the delivery as well as the voice, and let it choose its reference from the liveliest part of the recording rather than the cleanest. It compares the new read against the original at matched volume, checks pitch variation and emphasis rather than timbre alone, then fits the edit around where the new read breathes.
What do you type into Vizard Agent?
Describe the delivery in ordinary words. A phrase like "conversational, varied pitch, emphasis on the key words, pauses that mean something" gives Vizard Agent something specific to aim at and, more usefully still, something concrete to check the finished read against afterwards.
Prompt
What does Vizard Agent actually do?
A voice match has two halves. Timbre is what people mean by cloning, and it comes out well from any clean sample. Manner is pace, emphasis and where the pauses fall, and a clean sample teaches none of it. That's why the reference has to be the liveliest passage, not the calmest. In the session behind this article it:
- Checked that the reference for an advert was one clear voice, built the voice match and timed it to the existing scenes.
- Refined the read after the first attempt came back rushed and synthetic, then compared it with the reference at matched volume, since a louder read only sounds more confident.
- Checked how the person said a difficult name in the original recording and applied a phonetic guide without changing the script.
- Found the most expressive passages of the recording and re-read using that animated passage as the reference, after a technically accurate match was rejected for sounding like somebody reading aloud.
- Fitted the edit around the new pauses rather than stretching the read, three passes in all.
When does this not work well?
Manner is much harder to borrow than tone. A reference that's short, flat, or recorded with music underneath gives Vizard Agent very little to learn from, and a script written in nobody's particular voice will sound stilted whoever reads it. Fixed, immovable timings leave the pauses nowhere to go.
Common questions
Why does a clean sample sound flat? Because it is flat. Vizard Agent looks for an animated passage instead.
What about a name it gets wrong? Vizard Agent checks the original recording for how the person says it.
Will the timing still fit the picture? Vizard Agent adjusts the edit around the new pauses rather than compressing them.