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

Arousal Shift VoiceNet dimension guide

Temporal Dynamicscode vn_ARSHslot 70
This dimension measures the change in physiological activation over the clip. 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.

Arousal Shift controls the change in physiological activation over the clip. The __high LoRA shifts it +0.43 at 100% (worst cost: burst-blend -0.35); the __low LoRA shifts it -0.14 at 125% (worst cost: speech quality -0.34). The __high direction is the stronger, more reliable knob here.
▲ HIGH — Arousal Shift __high
Dose-response
dosetargetΔtargetΔgenuΔblendΔqual
baseline (0%)0.203————
25%0.228+0.02-0.02-0.02-0.03
50%0.327+0.12-0.03-0.20+0.06
75%0.459+0.26-0.05-0.21-0.04
100% ★0.637+0.43-0.10-0.35+0.01
125%0.634+0.43-0.09-0.37-0.13
EmoNet emotion correlation
Triumph+0.63
Pride+0.61
Anger+0.48
Relief-0.62
Affection-0.61
Longing-0.61
VoiceNet dimension correlation
Attack+0.89
Arousal+0.89
Stance+0.86
Vulnerability-0.83
Cognitive Load-0.74
Genuineness-0.67
Listen — 3 samples per dose
25%measured target: 0.33 · 0.29 · 0.19
50%measured target: 0.51 · 0.10 · 0.47
75%measured target: 0.38 · 0.56 · 0.52
100% ★measured target: 0.70 · 0.87 · 0.75
125%measured target: 0.56 · 0.58 · 0.61
Merging Arousal Shift __high raises the dimension: at the recommended 100% dose the measured Arousal Shift score moves from a 0.20 baseline to 0.64 (a +0.43 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 (-0.35); across the board genuineness -0.10, burst-blend -0.35, speech quality +0.01. Emotionally it reads as Triumph (r 0.63), Pride (0.61) and Anger (0.48), and it moves away from Relief, Affection and Longing. On other voice dimensions it co-moves most with Attack (0.89), Arousal (0.89) and Stance (0.86), while pulling down Vulnerability, Cognitive Load and Genuineness. Practical call: merge __high at 75–100%; expect Triumph and Attack to rise and burst-blend to fall. If you don’t want Triumph bleeding in, offset it with a prompt cue toward Relief to keep the arousal shift shift without the side-effect.
▼ LOW — Arousal Shift __low
Dose-response
dosetargetΔtargetΔgenuΔblendΔqual
baseline (0%)0.203————
25%0.211+0.01-0.01-0.04+0.00
50%0.145-0.06+0.02-0.09-0.11
75%0.123-0.08+0.07-0.02-0.22
100%0.118-0.08+0.04-0.08-0.14
125% ★0.064-0.14+0.07-0.11-0.34
EmoNet emotion correlation
Interest+0.72
Affection+0.61
Hope/Enthusiasm/Optimism+0.58
Pain-0.69
Fatigue/Exhaustion-0.64
Disappointment-0.56
VoiceNet dimension correlation
Structure+0.76
Smoothness+0.75
Fullness+0.73
Genuineness-0.64
Disfluency-0.59
Playful Style-0.46
Listen — 3 samples per dose
25%measured target: 0.22 · 0.30 · 0.18
50%measured target: 0.09 · 0.20 · 0.10
75%measured target: 0.05 · 0.20 · 0.00
100%measured target: 0.20 · 0.04 · 0.24
125% ★measured target: 0.06 · 0.00 · 0.10
Merging Arousal Shift __low damps the dimension: at the recommended 125% dose the measured Arousal Shift score moves from a 0.20 baseline to 0.06 (a -0.14 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 speech quality (-0.34); across the board genuineness +0.07, burst-blend -0.11, speech quality -0.34. Emotionally it reads as Interest (r 0.72), Affection (0.61) and Hope/Enthusiasm/Optimism (0.58), and it moves away from Pain, Fatigue/Exhaustion and Disappointment. On other voice dimensions it co-moves most with Structure (0.76), Smoothness (0.75) and Fullness (0.73), while pulling down Genuineness, Disfluency and Playful Style. Practical call: merge __low at 100–125%; expect Interest and Structure to rise and speech quality to fall. If you don’t want Interest bleeding in, offset it with a prompt cue toward Pain to keep the arousal shift 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