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.
aqir
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_MONO ▲ S_DRAM ▲ AGEV ▼ S_CONV ▼ S_NEWS ▼ EXPL ▼ DARC · emo: Malevolence_Malice Sexual_Lust Pride
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 103/518.
Best 5 of the kept top-20%
R 6.12 · genu 0.20 · blend 1.00 · qual 0.59 · WER 0.80
R 4.91 · genu 0.27 · blend 1.00 · qual 0.45 · WER 0.92
R 4.57 · genu 0.28 · blend 1.00 · qual 0.50 · WER 0.83
R 4.08 · genu 0.21 · blend 0.21 · qual 0.63 · WER 0.00
R 3.77 · genu 0.15 · blend 0.41 · qual 0.62 · WER 0.14
Weakest 5 still inside the top-20% (the selection boundary)
R 1.31 · genu 0.16 · blend 0.41 · qual 0.34 · WER 0.10
R 1.31 · genu 0.11 · blend 0.07 · qual 0.64 · WER 0.00
R 1.31 · genu 0.15 · blend 0.11 · qual 0.58 · WER 0.00
R 1.31 · genu 0.11 · blend 0.00 · qual 0.67 · WER 0.00
R 1.30 · genu 0.22 · blend 0.33 · qual 0.37 · WER 0.27
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.”