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.
crow
Characteristic dimensions (auto, from its own voice profile): ▲ S_STRY ▲ AROU ▲ REGS ▲ CHNK ▼ S_RANT ▼ EXPL ▼ COGL ▼ S_TECH · emo: Concentration Interest Contemplation
Training set = top-20% by reward = z(genu)+z(blend)+z(quality)−z(WER)+0.7·z(profile-match). Kept 128/640.
Best 5 of the kept top-20%
R 5.37 · genu 0.16 · blend 0.43 · qual 0.58 · WER 0.00
R 5.27 · genu 0.16 · blend 0.36 · qual 0.61 · WER 0.00
R 5.26 · genu 0.20 · blend 0.43 · qual 0.55 · WER 0.00
R 5.09 · genu 0.24 · blend 0.36 · qual 0.54 · WER 0.00
R 5.03 · genu 0.17 · blend 0.35 · qual 0.64 · WER 0.00
Weakest 5 still inside the top-20% (the selection boundary)
R 1.35 · genu 0.16 · blend 0.18 · qual 0.49 · WER 0.00
R 1.35 · genu 0.01 · blend 0.03 · qual 0.70 · WER 0.00
R 1.31 · genu 0.03 · blend 0.04 · qual 0.69 · WER 0.03
R 1.31 · genu 0.03 · blend 0.00 · qual 0.69 · WER 0.00
R 1.30 · genu 0.11 · blend 0.15 · 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.”