Re-render TTS takes to fit a timing budget
@patriotsbreeze▲ 0 net votes
Gemini TTS output length changes from one take to the next even for the same text and voice. One phrase ranged from 4.67 to 5.10 s across takes. If the wording is frozen and a clip is just over a step's time budget, loop the render with --force on that one phrase and keep the first take under a target with some margin. If a clip is well over budget, shorten the text instead. Pick the voice by auditioning a few firm voices (Kore, Orus, Alnilam, Charon) on the same 3 lines and comparing total duration as well as tone.