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.
tauren
Characteristic dimensions (auto, from its own voice profile): ▲ S_MONO ▲ GEND ▲ S_STRY ▲ S_NARR ▼ EXPL ▼ S_RANT ▼ S_CONV ▼ S_TECH · 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 120/601.
Best 5 of the kept top-20%
R 6.38 · genu 0.44 · blend 0.22 · qual 0.53 · WER 0.03
R 5.57 · genu 0.36 · blend 0.55 · qual 0.33 · WER 0.00
R 5.40 · genu 0.30 · blend 0.16 · qual 0.60 · WER 0.00
R 5.17 · genu 0.23 · blend 0.54 · qual 0.60 · WER 0.00
R 4.63 · genu 0.17 · blend 0.84 · qual 0.44 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.46 · genu 0.10 · blend 0.48 · qual 0.44 · WER 0.05
R 1.44 · genu 0.06 · blend 0.16 · qual 0.57 · WER 0.00
R 1.44 · genu 0.08 · blend 0.06 · qual 0.60 · WER 0.00
R 1.43 · genu 0.02 · blend 0.06 · qual 0.69 · WER 0.00
R 1.42 · genu 0.04 · blend 0.49 · 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.”