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.
vulpera
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_MONO ▲ S_NARR ▲ STRU ▼ S_RANT ▼ EXPL ▼ COGL ▼ DFLU · 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 127/637.
Best 5 of the kept top-20%
R 6.86 · genu 0.22 · blend 0.67 · qual 0.59 · WER 0.00
R 5.85 · genu 0.36 · blend 0.15 · qual 0.58 · WER 0.00
R 5.78 · genu 0.19 · blend 0.45 · qual 0.69 · WER 0.00
R 5.57 · genu 0.23 · blend 0.28 · qual 0.61 · WER 0.00
R 5.15 · genu 0.38 · blend 0.25 · qual 0.46 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.03 · genu 0.03 · blend 0.11 · qual 0.60 · WER 0.00
R 1.03 · genu 0.12 · blend 0.00 · qual 0.56 · WER 0.00
R 1.03 · genu 0.05 · blend 0.23 · qual 0.59 · WER 0.14
R 1.03 · genu 0.03 · blend 0.00 · qual 0.67 · WER 0.00
R 1.02 · genu 0.14 · blend 0.07 · qual 0.51 · 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.”