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.
gravelly-dark-titan
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_NARR ▲ S_MONO ▲ ESTH ▼ S_RANT ▼ EXPL ▼ S_TECH ▼ COGL · 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 128/640.
Best 5 of the kept top-20%
R 6.00 · genu 0.09 · blend 0.57 · qual 0.65 · WER 0.00
R 4.93 · genu 0.17 · blend 0.76 · qual 0.47 · WER 0.03
R 4.82 · genu 0.13 · blend 0.64 · qual 0.50 · WER 0.00
R 4.63 · genu 0.08 · blend 0.38 · qual 0.68 · WER 0.00
R 4.58 · genu 0.34 · blend 0.09 · qual 0.45 · WER 0.14
Weakest 5 still inside the top-20% (the selection boundary)
R 1.33 · genu 0.07 · blend 0.07 · qual 0.61 · WER 0.00
R 1.32 · genu 0.10 · blend 0.19 · qual 0.72 · WER 0.00
R 1.32 · genu 0.04 · blend 0.27 · qual 0.61 · WER 0.00
R 1.30 · genu 0.14 · blend 0.12 · qual 0.51 · WER 0.03
R 1.30 · genu 0.08 · blend 0.00 · qual 0.65 · 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.”