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.
murloc
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_MONO ▲ GEND ▲ S_NARR ▼ S_RANT ▼ EXPL ▼ S_ASMR ▼ DFLU · emo: Concentration Interest Emotional_Numbness
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 124/623.
Best 5 of the kept top-20%
R 10.32 · genu 0.68 · blend 0.42 · qual 0.49 · WER 0.00
R 9.19 · genu 0.63 · blend 0.47 · qual 0.42 · WER 0.00
R 6.54 · genu 0.42 · blend 0.67 · qual 0.33 · WER 0.00
R 5.95 · genu 0.38 · blend 0.47 · qual 0.44 · WER 0.00
R 5.60 · genu 0.44 · blend 0.25 · qual 0.50 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.00 · genu 0.01 · blend 0.01 · qual 0.68 · WER 0.00
R 1.00 · genu 0.05 · blend 0.00 · qual 0.65 · WER 0.00
R 0.98 · genu 0.12 · blend 0.04 · qual 0.56 · WER 0.03
R 0.98 · genu 0.23 · blend 0.19 · qual 0.29 · WER 0.07
R 0.97 · genu 0.03 · blend 0.00 · qual 0.67 · 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.”