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.
naga
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_MONO ▲ S_DRAM ▲ S_NARR ▼ S_RANT ▼ EXPL ▼ S_NEWS ▼ S_TECH · emo: Concentration Interest Awe
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 127/638.
Best 5 of the kept top-20%
R 5.95 · genu 0.15 · blend 0.36 · qual 0.70 · WER 0.00
R 5.44 · genu 0.11 · blend 0.30 · qual 0.73 · WER 0.00
R 4.67 · genu 0.11 · blend 0.39 · qual 0.75 · WER 0.00
R 4.62 · genu 0.19 · blend 0.17 · qual 0.62 · WER 0.00
R 4.61 · genu 0.17 · blend 0.19 · qual 0.69 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.28 · genu 0.12 · blend 0.02 · qual 0.58 · WER 0.00
R 1.28 · genu 0.00 · blend 0.01 · qual 0.77 · WER 0.00
R 1.27 · genu 0.04 · blend 0.07 · qual 0.62 · WER 0.00
R 1.25 · genu 0.11 · blend 0.00 · qual 0.61 · WER 0.03
R 1.24 · genu 0.13 · blend 0.28 · qual 0.56 · 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.”