← all dimensions · markdown (agent skill) · manual home

Stance VoiceNet dimension guide

Social & Interpersonalcode vn_STNCslot 48
This dimension measures the speaker’s social posture in the conversational space. Below: the merge-scaling dose-response for the __high and __low LoRA (laion/moss-voicenet-dimension-loras) at 25/50/75/100/125%, the measured target value at each dose, the top-3 ± correlated EmoNet emotions and VoiceNet dimensions, three audio samples per dose, and a grounded prompting tip. Costs are shifts vs. the 24-sample neutral baseline in genuineness, vocal-burst blend and speech quality.

Stance controls the speaker’s social posture in the conversational space. The __high LoRA shifts it +0.58 at 125% (worst cost: burst-blend -0.38); the __low LoRA shifts it -0.12 at 125% (worst cost: burst-blend -0.12). The __high direction is the stronger, more reliable knob here.
▲ HIGH — Stance __high
Dose-response
dosetargetΔtargetΔgenuΔblendΔqual
baseline (0%)0.269————
25%0.376+0.11-0.06-0.03+0.08
50%0.539+0.27-0.06-0.30-0.01
75%0.681+0.41-0.09-0.37-0.04
100%0.757+0.49-0.10-0.40-0.08
125% ★0.846+0.58-0.07-0.38-0.20
EmoNet emotion correlation
Triumph+0.72
Anger+0.62
Pride+0.49
Contemplation-0.72
Contentment-0.69
Relief-0.62
VoiceNet dimension correlation
Tension+0.96
Attack+0.96
Metallic Character+0.96
Vulnerability-0.88
Warmth-0.83
Whisper-Talk Style-0.80
Listen — 3 samples per dose
25%measured target: 0.22 · 0.26 · 0.45
50%measured target: 0.59 · 0.64 · 0.17
75%measured target: 0.71 · 0.54 · 0.70
100%measured target: 0.51 · 0.84 · 0.59
125% ★measured target: 0.95 · 0.78 · 0.84
Merging Stance __high pushes the dimension: at the recommended 125% dose the measured Stance score moves from a 0.27 baseline to 0.85 (a +0.58 shift). The dose-response is clean and monotonic — each step pushes the dimension further. That reach is a steep cost — the biggest hit is to burst-blend (-0.38); across the board genuineness -0.07, burst-blend -0.38, speech quality -0.20. Emotionally it reads as Triumph (r 0.72), Anger (0.62) and Pride (0.49), and it moves away from Contemplation, Contentment and Relief. On other voice dimensions it co-moves most with Tension (0.96), Attack (0.96) and Metallic Character (0.96), while pulling down Vulnerability, Warmth and Whisper-Talk Style. Practical call: merge __high at 100–125%; expect Triumph and Tension to rise and burst-blend to fall. If you don’t want Triumph bleeding in, offset it with a prompt cue toward Contemplation to keep the stance shift without the side-effect.
▼ LOW — Stance __low
Dose-response
dosetargetΔtargetΔgenuΔblendΔqual
baseline (0%)0.269————
25%0.207-0.06+0.01+0.07+0.06
50%0.257-0.01+0.01+0.06+0.00
75%0.201-0.07-0.01-0.01+0.08
100%0.193-0.08-0.05+0.02+0.10
125% ★0.151-0.12-0.06-0.12+0.16
EmoNet emotion correlation
Jealousy & Envy+0.39
Malevolence/Malice+0.34
Impatience and Irritability+0.30
Contemplation-0.46
Concentration-0.46
Emotional Numbness-0.42
VoiceNet dimension correlation
Tension+0.79
Emphasis+0.79
Attack+0.75
Teacher/Didactic Style-0.64
ASMR Style-0.61
Register-0.61
Listen — 3 samples per dose
25%measured target: 0.12 · 0.13 · 0.16
50%measured target: 0.26 · 0.54 · 0.32
75%measured target: 0.19 · 0.29 · 0.21
100%measured target: 0.22 · 0.32 · 0.13
125% ★measured target: 0.09 · 0.15 · 0.17
Merging Stance __low lowers the dimension: at the recommended 125% dose the measured Stance score moves from a 0.27 baseline to 0.15 (a -0.12 shift). The dose-response is clean and monotonic — each step pushes the dimension further. That reach is a modest cost — the biggest hit is to burst-blend (-0.12); across the board genuineness -0.06, burst-blend -0.12, speech quality +0.16. Emotionally it reads as Jealousy & Envy (r 0.39), Malevolence/Malice (0.34) and Impatience and Irritability (0.30), and it moves away from Contemplation, Concentration and Emotional Numbness. On other voice dimensions it co-moves most with Tension (0.79), Emphasis (0.79) and Attack (0.75), while pulling down Teacher/Didactic Style, ASMR Style and Register. Practical call: merge __low at 100–125%; expect Jealousy & Envy and Tension to rise and burst-blend to fall. If you don’t want Jealousy & Envy bleeding in, offset it with a prompt cue toward Contemplation to keep the stance shift without the side-effect.
Data: MOSS-VA-v2 base + per-dimension LoRA, neutral text, no reference · 5 merge doses × 8 scored samples/dose · correlations are Pearson r over the sample pool · scaling grid · LoRA repo