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.
voice-12
Characteristic dimensions (auto, from its own voice profile): ▲ STRU ▲ S_NARR ▲ S_STRY ▲ S_MONO ▼ EXPL ▼ S_RANT ▼ DARC ▼ S_CART · emo: Concentration Interest Awe
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 9.05 · genu 0.25 · blend 0.60 · qual 0.57 · WER 0.08
R 4.88 · genu 0.12 · blend 0.24 · qual 0.63 · WER 0.00
R 4.70 · genu 0.18 · blend 0.14 · qual 0.66 · WER 0.00
R 4.48 · genu 0.16 · blend 0.50 · qual 0.71 · WER 0.70
R 4.37 · genu 0.13 · blend 0.25 · qual 0.68 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.13 · genu 0.04 · blend 0.15 · qual 0.65 · WER 0.10
R 1.12 · genu 0.14 · blend 0.08 · qual 0.61 · WER 0.05
R 1.12 · genu 0.05 · blend 0.00 · qual 0.66 · WER 0.00
R 1.11 · genu 0.05 · blend 0.41 · qual 0.73 · WER 0.55
R 1.11 · genu 0.00 · blend 0.01 · qual 0.70 · 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.”