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.
patch
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ TEMP ▲ EMPH ▲ STNC ▼ VULN ▼ S_ASMR ▼ S_WHIS ▼ S_TECH · emo: Amusement Teasing Malevolence_Malice
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 126/632.
Best 5 of the kept top-20%
R 5.00 · genu 0.13 · blend 0.30 · qual 0.62 · WER 0.09
R 4.71 · genu 0.18 · blend 0.24 · qual 0.57 · WER 0.00
R 4.67 · genu 0.27 · blend 0.35 · qual 0.43 · WER 0.29
R 4.55 · genu 0.19 · blend 0.38 · qual 0.47 · WER 0.00
R 4.39 · genu 0.23 · blend 0.28 · qual 0.44 · WER 0.07
Weakest 5 still inside the top-20% (the selection boundary)
R 1.21 · genu 0.16 · blend 0.29 · qual 0.40 · WER 0.03
R 1.18 · genu 0.12 · blend 0.25 · qual 0.27 · WER 0.00
R 1.17 · genu 0.09 · blend 0.24 · qual 0.34 · WER 0.03
R 1.16 · genu 0.10 · blend 0.36 · qual 0.26 · WER 0.00
R 1.15 · genu 0.05 · blend 0.13 · qual 0.50 · WER 0.03
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.”