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.
high_elf
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_MONO ▲ ESTH ▲ RCQL ▼ S_RANT ▼ EXPL ▼ S_TECH ▼ DARC · emo: Concentration Interest Awe
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 7.16 · genu 0.26 · blend 0.53 · qual 0.56 · WER 0.00
R 5.69 · genu 0.21 · blend 0.43 · qual 0.46 · WER 0.00
R 5.57 · genu 0.20 · blend 0.36 · qual 0.60 · WER 0.00
R 5.22 · genu 0.21 · blend 0.53 · qual 0.48 · WER 0.00
R 5.17 · genu 0.13 · blend 0.25 · qual 0.70 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.46 · genu 0.01 · blend 0.06 · qual 0.69 · WER 0.00
R 1.45 · genu 0.01 · blend 0.06 · qual 0.71 · WER 0.00
R 1.45 · genu 0.14 · blend 0.12 · qual 0.52 · WER 0.00
R 1.44 · genu 0.07 · blend 0.25 · qual 0.57 · WER 0.00
R 1.41 · genu 0.10 · blend 0.05 · qual 0.60 · 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.”