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.
voice-16
Characteristic dimensions (auto, from its own voice profile): ▲ S_MONO ▲ S_STRY ▲ GEND ▲ FULL ▼ EXPL ▼ REGS ▼ S_CONV ▼ S_TECH · emo: Concentration Emotional_Numbness Interest
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 126/631.
Best 5 of the kept top-20%
R 5.97 · genu 0.09 · blend 0.49 · qual 0.68 · WER 0.04
R 5.42 · genu 0.16 · blend 0.85 · qual 0.60 · WER 0.74
R 4.84 · genu 0.08 · blend 0.39 · qual 0.67 · WER 0.00
R 4.72 · genu 0.09 · blend 0.45 · qual 0.59 · WER 0.00
R 4.66 · genu 0.20 · blend 0.81 · qual 0.58 · WER 0.84
Weakest 5 still inside the top-20% (the selection boundary)
R 1.31 · genu 0.09 · blend 0.00 · qual 0.54 · WER 0.03
R 1.31 · genu 0.03 · blend 0.06 · qual 0.61 · WER 0.03
R 1.29 · genu 0.11 · blend 0.31 · qual 0.34 · WER 0.00
R 1.28 · genu 0.00 · blend 0.13 · qual 0.59 · WER 0.00
R 1.27 · genu 0.08 · blend 0.07 · qual 0.56 · WER 0.09
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.”