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.
nightborne_elf
Characteristic dimensions (auto, from its own voice profile): ▲ S_MONO ▲ S_STRY ▲ ESTH ▲ STRU ▼ S_RANT ▼ EXPL ▼ DARC ▼ S_TECH · emo: Concentration Emotional_Numbness Interest
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 127/639.
Best 5 of the kept top-20%
R 7.07 · genu 0.18 · blend 0.50 · qual 0.68 · WER 0.00
R 5.77 · genu 0.20 · blend 0.46 · qual 0.65 · WER 0.00
R 5.43 · genu 0.11 · blend 0.51 · qual 0.62 · WER 0.00
R 5.39 · genu 0.11 · blend 0.43 · qual 0.68 · WER 0.00
R 5.04 · genu 0.16 · blend 0.49 · qual 0.63 · WER 0.11
Weakest 5 still inside the top-20% (the selection boundary)
R 1.54 · genu 0.08 · blend 0.05 · qual 0.60 · WER 0.03
R 1.51 · genu 0.21 · blend 0.01 · qual 0.56 · WER 0.00
R 1.50 · genu 0.06 · blend 0.18 · qual 0.65 · WER 0.00
R 1.49 · genu 0.04 · blend 0.11 · qual 0.70 · WER 0.00
R 1.49 · genu 0.00 · blend 0.19 · qual 0.71 · 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.”