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.
orc
Characteristic dimensions (auto, from its own voice profile): ▲ S_MONO ▲ GEND ▲ S_STRY ▲ S_NARR ▼ S_RANT ▼ S_ASMR ▼ EXPL ▼ S_TECH · emo: Concentration Interest Emotional_Numbness
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 8.30 · genu 0.41 · blend 0.29 · qual 0.59 · WER 0.00
R 7.16 · genu 0.10 · blend 0.59 · qual 0.68 · WER 0.00
R 6.24 · genu 0.14 · blend 0.63 · qual 0.50 · WER 0.04
R 5.89 · genu 0.12 · blend 0.56 · qual 0.64 · WER 0.04
R 5.34 · genu 0.05 · blend 0.28 · qual 0.78 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.23 · genu 0.01 · blend 0.10 · qual 0.59 · WER 0.00
R 1.22 · genu 0.03 · blend 0.04 · qual 0.64 · WER 0.00
R 1.22 · genu 0.08 · blend 0.25 · qual 0.51 · WER 0.00
R 1.19 · genu 0.12 · blend 0.19 · qual 0.68 · WER 0.13
R 1.18 · genu 0.08 · blend 0.04 · qual 0.53 · 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.”