Why voices drift
Three things make a voice sound different from one line to the next:
- A different sample. The sample is the voice. Replace it and every new take sounds different.
- Different settings. Moving the sliders between sessions changes the delivery of everything made afterwards.
- Natural variation. Every take is a fresh performance, a little different each time, just like a human actor's.
One sample per voice, for the whole production
Record a voice's sample once, carefully, and keep it. If a character has to change, for example because a better recording comes along, decide whether it's worth it: lines you've saved keep the clip they were made with, so the finished episodes won't change, but new lines will sound like the new clip, and the two may not match side by side. Early in a project, replacing a clip is cheap; late in it, it can mean re-voicing.
The same goes across projects. Each project has its own cast, so use the same WAV in each: Download clip on the voice, then upload that file to the other project's cast.
Set the style once
Each voice keeps its own style and slider positions, saved the next time you generate. Tune a voice at the start with a few test lines, then leave it. If you're tempted to change it later for a single line, generate another take of that line instead, or give the line to a second voice made for that mood.
Keep variation in check
- Expressiveness controls how much takes differ. For a narrator who has to sound identical across fifty modules, keep it around 0.6 to 0.8.
- Calm gives the steadiest delivery, and suits long-form narration.
- Similar line lengths help. A two-word line and a two-minute paragraph get read differently. One paragraph per line keeps narration even.
- Don't mix models. Fast draft uses a different, quicker model. It's great for drafts, but its takes won't match the others, so re-generate final lines in the voice's main style.
Save lines as you go
When a line sounds right, save it. A saved line is locked: it keeps the take, the text it says, the voice that read it and the clip that voice was cloned from, even if the voice is later renamed, re-recorded or deleted. Saving is what makes a project safe to keep working on. To change a saved line, press Edit on it.
Check in context
Consistency is easiest to hear in sequence. Before saving a new line in an old script, play the line before and after it. In a big production, keep a "reference" line for each voice, one you know is right, and compare new takes with it.
Team projects
When several people can edit a project, agree who owns each voice's settings, and note them in the project's comments. One person changing a narrator's sliders affects every line anyone generates afterwards. See collaborating on a project.