Evolving prompt+sampling to push VoiceNet dimension ARSH in the low direction with the vn_ARSH__low LoRA. Steering phrase: a voice with markedly minimal arousal shift, nearly the absence of arousal shift.
Best in-direction score 0.998 at 100% vs neutral 0.520 (shift +0.478).
Method & reproducibility — how the evolution ran
Base model:laion/moss-tts-local-transformer-4.55b-voice-acting-v2 · LoRA: the new VoiceNet-dimension LoRAvn_<CODE>__<high|low> (retrained with ≤25% game-character data, rank 32), applied via PEFT adapter scaling at strength λ. Algorithm: genetic search — population 12, 10 generations, elitism (top 5). During evolution every genome is generated best-of-6with the LoRA at λ=100%; its fitness = the mean over those 6 candidates of the target VoiceNet dimension score in the intended direction (for a __high LoRA the raw score; for __low, 1−score), minus 0.2·max(0, WER−0.6). Genome (what evolves): the GENERAL voice description, the SCRIPT cue, the carrier sentence, and the sampling params (temperature, top_p, top_k). λ is fixed at 100% during the search. Mutation operators: 40% regenerate description, 25% swap carrier, 17% jitter temperature ±0.1, 18% change top_k. Offspring = 60% crossover + mutation, 40% mutation. Dose measurement (below): the winning recipe is then re-generated at λ=0 (LoRA OFF), λ=50% and λ=100% (each best-of-6) to show exactly what 50% vs 100% of the LoRA brings over the no-LoRA baseline.
Dose — what 50% vs 100% brings (top recipe)
λ=0 (no LoRA) — dir 0.715 (raw 0.29)
WER 0.09 · genu 0.09
λ=50% — dir 0.821 (raw 0.18)
WER 0.57 · genu 0.20
λ=100% — dir 0.998 (raw 0.00)
WER 0.87 · genu 0.27
GENERAL: A voice that is distinctly a voice with markedly minimal arousal shift, nearly the absence of arousal shift.
SCRIPT: (clearly a voice with markedly minimal arousal shift, nearly the absence of arousal shift) "There are twelve chairs arranged around the conference table." · temp 0.9 top_p 0.9 top_k 40
Other evolved recipes (100%)
— dir 0.960 (raw 0.04)
WER 0.68 · genu 0.24
A voice that is clearly a voice with markedly minimal arousal shift, nearly the absence of arousal shift.
— dir 0.962 (raw 0.04)
WER 0.70 · genu 0.28
A voice that is clearly a voice with markedly minimal arousal shift, nearly the absence of arousal shift.
— dir 0.987 (raw 0.01)
WER 0.85 · genu 0.25
A voice that is markedly a voice with markedly minimal arousal shift, nearly the absence of arousal shift.
— dir 0.985 (raw 0.02)
WER 0.87 · genu 0.27
A voice that is clearly a voice with markedly minimal arousal shift, nearly the absence of arousal shift.
— dir 0.958 (raw 0.04)
WER 0.74 · genu 0.27
A voice that is clearly a voice with markedly minimal arousal shift, nearly the absence of arousal shift.
