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.
jinyu
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_MONO ▲ S_NARR ▲ GEND ▼ S_RANT ▼ EXPL ▼ S_ASMR ▼ DARC · emo: Concentration Interest Awe
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 118/593.
Best 5 of the kept top-20%
R 5.86 · genu 0.17 · blend 0.67 · qual 0.55 · WER 0.04
R 4.83 · genu 0.18 · blend 0.67 · qual 0.58 · WER 0.03
R 4.71 · genu 0.18 · blend 0.74 · qual 0.47 · WER 0.07
R 4.60 · genu 0.13 · blend 0.59 · qual 0.66 · WER 0.00
R 4.28 · genu 0.12 · blend 0.47 · qual 0.62 · WER 0.03
Weakest 5 still inside the top-20% (the selection boundary)
R 1.52 · genu 0.05 · blend 0.18 · qual 0.63 · WER 0.00
R 1.52 · genu 0.08 · blend 0.00 · qual 0.74 · WER 0.03
R 1.47 · genu 0.15 · blend 0.19 · qual 0.60 · WER 0.08
R 1.46 · genu 0.11 · blend 0.10 · qual 0.55 · WER 0.00
R 1.46 · genu 0.11 · blend 0.41 · qual 0.50 · 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.”