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.
fire_elemental
Characteristic dimensions (auto, from its own voice profile): ▲ GEND ▲ S_STRY ▲ S_MONO ▲ S_NARR ▼ EXPL ▼ S_RANT ▼ DARC ▼ S_CONV · 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 128/640.
Best 5 of the kept top-20%
R 5.98 · genu 0.18 · blend 0.52 · qual 0.62 · WER 0.00
R 5.84 · genu 0.13 · blend 0.49 · qual 0.65 · WER 0.00
R 5.55 · genu 0.14 · blend 0.55 · qual 0.57 · WER 0.00
R 5.52 · genu 0.16 · blend 0.69 · qual 0.46 · WER 0.00
R 5.27 · genu 0.15 · blend 0.47 · qual 0.62 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.69 · genu 0.09 · blend 0.07 · qual 0.61 · WER 0.00
R 1.67 · genu 0.09 · blend 0.07 · qual 0.57 · WER 0.00
R 1.66 · genu 0.06 · blend 0.11 · qual 0.61 · WER 0.00
R 1.65 · genu 0.12 · blend 0.08 · qual 0.57 · WER 0.00
R 1.64 · genu 0.00 · blend 0.16 · qual 0.72 · 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.”