Why it happens
- Different voices or clone sources were picked for different lines.
- Emotion settings are pushed so hard they drag age and pitch along with them.
- The voice model or version changed between episodes.
How to fix it
Write a voice spec per character: perceived age, pitch, pace, breathiness, emotional range and things to avoid.
Compare candidate voices using the same test line, then record the chosen voice ID and settings.
Carry emotion through the line, pauses and pace, and adjust emotion strength only within a narrow range.
Play all of a character's lines back to back on one track, find the odd one out and redo only that line.
Prevent it next time：Set a voice sheet with IDs and settings before recording any dialogue, and reuse it for the whole series.