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-04
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ S_MONO ▲ ESTH ▲ RCQL ▼ S_RANT ▼ S_TECH ▼ EXPL ▼ S_ASMR · emo: Concentration Interest Awe
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 124/621.
Best 5 of the kept top-20%
R 6.04 · genu 0.18 · blend 0.52 · qual 0.61 · WER 0.00
R 5.69 · genu 0.05 · blend 0.49 · qual 0.73 · WER 0.00
R 5.05 · genu 0.30 · blend 0.52 · qual 0.43 · WER 0.03
R 4.70 · genu 0.13 · blend 0.66 · qual 0.51 · WER 0.00
R 4.65 · genu 0.12 · blend 0.38 · qual 0.61 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.44 · genu 0.00 · blend 0.09 · qual 0.73 · WER 0.00
R 1.44 · genu 0.11 · blend 0.13 · qual 0.65 · WER 0.03
R 1.43 · genu 0.17 · blend 0.26 · qual 0.51 · WER 0.04
R 1.35 · genu 0.04 · blend 0.00 · qual 0.71 · WER 0.00
R 1.33 · genu 0.08 · blend 0.08 · qual 0.56 · 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.”