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.
earth_elemental
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ GEND ▲ S_MONO ▲ R_CHST ▼ REGS ▼ EXPL ▼ S_CONV ▼ R_MASK · emo: Concentration Malevolence_Malice Pride
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 121/605.
Best 5 of the kept top-20%
R 6.56 · genu 0.09 · blend 0.64 · qual 0.59 · WER 0.00
R 5.85 · genu 0.15 · blend 0.55 · qual 0.47 · WER 0.00
R 5.11 · genu 0.14 · blend 0.53 · qual 0.59 · WER 0.00
R 4.54 · genu 0.10 · blend 0.52 · qual 0.52 · WER 0.00
R 4.44 · genu 0.13 · blend 0.38 · qual 0.57 · WER 0.07
Weakest 5 still inside the top-20% (the selection boundary)
R 1.57 · genu 0.09 · blend 0.17 · qual 0.56 · WER 0.00
R 1.57 · genu 0.09 · blend 0.12 · qual 0.56 · WER 0.00
R 1.57 · genu 0.10 · blend 0.23 · qual 0.53 · WER 0.03
R 1.52 · genu 0.04 · blend 0.37 · qual 0.67 · WER 0.41
R 1.50 · genu 0.10 · blend 0.24 · qual 0.46 · 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.”