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.
eredar
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_MONO ▲ GEND ▲ S_DRAM ▼ S_TECH ▼ EXPL ▼ S_CONV ▼ S_RANT · emo: Concentration Interest Awe
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 123/616.
Best 5 of the kept top-20%
R 6.02 · genu 0.35 · blend 0.82 · qual 0.25 · WER 0.03
R 5.76 · genu 0.21 · blend 0.77 · qual 0.49 · WER 0.00
R 4.85 · genu 0.20 · blend 0.49 · qual 0.51 · WER 0.00
R 4.42 · genu 0.10 · blend 0.78 · qual 0.56 · WER 0.00
R 3.84 · genu 0.16 · blend 0.34 · qual 0.60 · WER 0.08
Weakest 5 still inside the top-20% (the selection boundary)
R 1.53 · genu 0.17 · blend 0.13 · qual 0.50 · WER 0.00
R 1.51 · genu 0.17 · blend 0.23 · qual 0.41 · WER 0.00
R 1.50 · genu 0.13 · blend 0.06 · qual 0.50 · WER 0.03
R 1.47 · genu 0.03 · blend 0.27 · qual 0.56 · WER 0.03
R 1.45 · genu 0.08 · blend 0.26 · qual 0.49 · 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.”