Profile-reward recipe: generate 640 (batched) → ASR-filter+trim → SIDON → keep the top 20% by a per-character quality+profile reward → warm-start FT 3 epochs. Below: the quality range of the selected training set, then example generations after each epoch.
zuni
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_DRAM ▲ EMPH ▲ AROU ▼ S_TECH ▼ S_ASMR ▼ S_WHIS ▼ S_NEWS · emo: Concentration Interest Teasing
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 126/634.
Best 5 of the kept top-20%
R 7.31 · genu 0.13 · blend 0.52 · qual 0.49 · WER 0.00
R 6.02 · genu 0.16 · blend 0.28 · qual 0.49 · WER 0.00
R 4.99 · genu 0.27 · blend 0.71 · qual 0.31 · WER 0.82
R 4.77 · genu 0.20 · blend 0.10 · qual 0.47 · WER 0.03
R 4.66 · genu 0.15 · blend 0.23 · qual 0.43 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.52 · genu 0.11 · blend 0.05 · qual 0.45 · WER 0.07
R 1.51 · genu 0.03 · blend 0.13 · qual 0.48 · WER 0.00
R 1.51 · genu 0.09 · blend 0.26 · qual 0.31 · WER 0.03
R 1.50 · genu 0.15 · blend 0.36 · qual 0.36 · WER 0.52
R 1.50 · genu 0.05 · blend 0.00 · qual 0.55 · WER 0.00
Fantasy line — refined ep1 / ep2 / ep3
“The old road winds past the ruined tower where the last dragon knights once kept their watch.”
Sci-fi line — refined ep1 / ep2 / ep3
“The station drifted in low orbit while the reactor cooled and the scanners swept the empty dark.”
Neutral line — refined ep1 / ep2 / ep3
“The meeting is scheduled for three o'clock and the report is on the desk by the printer.”