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.
undead
Characteristic dimensions (auto, from its own voice profile): ▲ S_MONO ▲ S_STRY ▲ S_NARR ▲ STRU ▼ S_RANT ▼ EXPL ▼ DARC ▼ 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 127/635.
Best 5 of the kept top-20%
R 7.40 · genu 0.28 · blend 0.62 · qual 0.51 · WER 0.00
R 7.02 · genu 0.36 · blend 0.42 · qual 0.47 · WER 0.10
R 6.67 · genu 0.19 · blend 0.70 · qual 0.54 · WER 0.00
R 5.43 · genu 0.16 · blend 0.35 · qual 0.72 · WER 0.04
R 5.35 · genu 0.12 · blend 0.41 · qual 0.67 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.40 · genu 0.00 · blend 0.02 · qual 0.77 · WER 0.00
R 1.39 · genu 0.11 · blend 0.10 · qual 0.58 · WER 0.00
R 1.35 · genu 0.00 · blend 0.23 · qual 0.61 · WER 0.00
R 1.35 · genu 0.01 · blend 0.00 · qual 0.75 · WER 0.00
R 1.31 · genu 0.01 · blend 0.05 · qual 0.75 · 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.”