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-27
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_MONO ▲ S_DRAM ▲ S_CART ▼ EXPL ▼ S_CONV ▼ S_TECH ▼ S_NEWS · emo: Concentration Interest Awe
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 127/635.
Best 5 of the kept top-20%
R 5.57 · genu 0.19 · blend 0.87 · qual 0.55 · WER 0.74
R 5.01 · genu 0.13 · blend 0.80 · qual 0.59 · WER 0.54
R 4.89 · genu 0.11 · blend 0.71 · qual 0.48 · WER 0.07
R 4.63 · genu 0.31 · blend 0.28 · qual 0.36 · WER 0.00
R 4.34 · genu 0.22 · blend 0.34 · qual 0.57 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.40 · genu 0.18 · blend 0.12 · qual 0.51 · WER 0.00
R 1.39 · genu 0.15 · blend 0.60 · qual 0.45 · WER 0.70
R 1.38 · genu 0.13 · blend 0.03 · qual 0.66 · WER 0.07
R 1.36 · genu 0.10 · blend 0.17 · qual 0.40 · WER 0.00
R 1.33 · genu 0.15 · blend 0.00 · qual 0.53 · 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.”