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.
vexiona
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_DRAM ▲ S_MONO ▲ ESTH ▼ S_TECH ▼ EXPL ▼ S_RANT ▼ S_NEWS · 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 127/637.
Best 5 of the kept top-20%
R 6.35 · genu 0.14 · blend 0.36 · qual 0.53 · WER 0.00
R 6.28 · genu 0.12 · blend 0.38 · qual 0.55 · WER 0.00
R 6.00 · genu 0.09 · blend 0.46 · qual 0.56 · WER 0.00
R 5.76 · genu 0.12 · blend 0.31 · qual 0.68 · WER 0.00
R 5.71 · genu 0.19 · blend 0.32 · qual 0.47 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.56 · genu 0.03 · blend 0.02 · qual 0.67 · WER 0.00
R 1.56 · genu 0.12 · blend 0.00 · qual 0.56 · WER 0.00
R 1.55 · genu 0.12 · blend 0.10 · qual 0.50 · WER 0.00
R 1.54 · genu 0.10 · blend 0.08 · qual 0.51 · WER 0.00
R 1.54 · genu 0.13 · blend 0.00 · 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.”