Evolving prompt+sampling to push VoiceNet dimension RCQL in the low direction with the vn_RCQL__low LoRA. Steering phrase: a voice with markedly minimal recording quality, nearly the absence of recording quality.
Best in-direction score 0.794 at 100% vs neutral 0.303 (shift +0.491).
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 0.350 (raw 0.65)
WER 0.11 · genu 0.04
λ=50% — dir 0.326 (raw 0.67)
WER 0.22 · genu 0.09
λ=100% — dir 0.794 (raw 0.21)
WER 0.72 · genu 0.29
GENERAL: A voice that is markedly a voice with markedly minimal recording quality, nearly the absence of recording quality.
SCRIPT: (distinctly a voice with markedly minimal recording quality, nearly the absence of recording quality) "There are twelve chairs arranged around the conference table." · temp 0.9 top_p 0.95 top_k 30
Other evolved recipes (100%)
— dir 0.790 (raw 0.21)
WER 0.71 · genu 0.27
A voice that is extremely a voice with markedly minimal recording quality, nearly the absence of recording quality.
— dir 0.755 (raw 0.24)
WER 0.58 · genu 0.29
A voice that is extremely a voice with markedly minimal recording quality, nearly the absence of recording quality.
— dir 0.832 (raw 0.17)
WER 0.99 · genu 0.28
A voice that is extremely a voice with markedly minimal recording quality, nearly the absence of recording quality.
— dir 0.802 (raw 0.20)
WER 0.85 · genu 0.29
A voice that is intensely a voice with markedly minimal recording quality, nearly the absence of recording quality.
— dir 0.804 (raw 0.20)
WER 0.89 · genu 0.24
A voice that is distinctly a voice with markedly minimal recording quality, nearly the absence of recording quality.
