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.
satyr
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ GEND ▲ S_MONO ▲ ROUG ▼ REGS ▼ EXPL ▼ S_CONV ▼ R_MASK · emo: Concentration Pride Malevolence_Malice
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 126/630.
Best 5 of the kept top-20%
R 5.97 · genu 0.46 · blend 0.45 · qual 0.21 · WER 0.16
R 4.75 · genu 0.20 · blend 0.69 · qual 0.31 · WER 0.00
R 4.40 · genu 0.15 · blend 0.74 · qual 0.49 · WER 0.32
R 4.37 · genu 0.13 · blend 0.75 · qual 0.30 · WER 0.07
R 4.31 · genu 0.12 · blend 0.55 · qual 0.57 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.27 · genu 0.17 · blend 0.09 · qual 0.29 · WER 0.03
R 1.27 · genu 0.14 · blend 0.35 · qual 0.35 · WER 0.04
R 1.25 · genu 0.15 · blend 0.33 · qual 0.27 · WER 0.10
R 1.25 · genu 0.10 · blend 0.27 · qual 0.28 · WER 0.00
R 1.24 · genu 0.09 · blend 0.10 · qual 0.44 · 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.”