← all dimensions · markdown (agent skill) · manual home
Narrator Style VoiceNet dimension guide
Speaking Stylecode vn_S_NARRslot 86
This dimension measures how much the voice reads like a narrator. 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.
Narrator Style controls how much the voice reads like a narrator. The __high LoRA shifts it +0.39 at 125% (worst cost: burst-blend -0.25); the __low LoRA shifts it -0.06 at 125% (worst cost: burst-blend -0.34). The __high direction is the stronger, more reliable knob here.
▲ HIGH — Narrator Style __high
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.454 | — | — | — | — |
| 25% | 0.481 | +0.03 | -0.02 | +0.09 | +0.04 |
| 50% | 0.654 | +0.20 | -0.06 | -0.03 | +0.14 |
| 75% | 0.721 | +0.27 | -0.11 | -0.22 | +0.14 |
| 100% | 0.840 | +0.39 | -0.10 | -0.27 | +0.18 |
| 125% ★ | 0.846 | +0.39 | -0.10 | -0.25 | +0.18 |
EmoNet emotion correlation
Sexual Lust+0.37
Contemplation+0.34
Amusement+0.31
Relief-0.66
Fatigue/Exhaustion-0.43
Helplessness-0.42
VoiceNet dimension correlation
Formal Style+0.89
Smoothness+0.88
Structure+0.85
Casual Style-0.81
Disfluency-0.73
Genuineness-0.67
Listen — 3 samples per dose
25%measured target: 0.46 · 0.46 · 0.44
50%measured target: 0.70 · 0.70 · 0.82
75%measured target: 0.94 · 0.50 · 0.69
100%measured target: 0.93 · 0.74 · 0.95
125% ★measured target: 0.83 · 0.97 · 0.79
Merging Narrator Style __high pushes the dimension: at the recommended 125% dose the measured Narrator Style score moves from a 0.45 baseline to 0.85 (a +0.39 shift). The dose-response is clean and monotonic — each step pushes the dimension further. That reach is a real cost — the biggest hit is to burst-blend (-0.25); across the board genuineness -0.10, burst-blend -0.25, speech quality +0.18. Emotionally it reads as Sexual Lust (r 0.37), Contemplation (0.34) and Amusement (0.31), and it moves away from Relief, Fatigue/Exhaustion and Helplessness. On other voice dimensions it co-moves most with Formal Style (0.89), Smoothness (0.88) and Structure (0.85), while pulling down Casual Style, Disfluency and Genuineness. Practical call: merge __high at 100–125%; expect Sexual Lust and Formal Style to rise and burst-blend to fall. If you don’t want Sexual Lust bleeding in, offset it with a prompt cue toward Relief to keep the narrator style shift without the side-effect.
▼ LOW — Narrator Style __low
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.454 | — | — | — | — |
| 25% | 0.429 | -0.03 | -0.04 | -0.12 | -0.02 |
| 50% | 0.517 | +0.06 | -0.02 | -0.21 | -0.08 |
| 75% | 0.539 | +0.08 | -0.05 | -0.27 | -0.02 |
| 100% | 0.572 | +0.12 | -0.08 | -0.33 | -0.01 |
| 125% ★ | 0.392 | -0.06 | -0.04 | -0.34 | -0.14 |
EmoNet emotion correlation
Interest+0.53
Longing+0.47
Contemplation+0.43
Pain-0.57
Distress-0.55
Fatigue/Exhaustion-0.50
VoiceNet dimension correlation
Smoothness+0.85
Esthetics+0.83
Structure+0.81
Disfluency-0.79
Genuineness-0.58
Casual Style-0.53
Listen — 3 samples per dose
25%measured target: 0.07 · 0.36 · 0.46
50%measured target: 0.87 · 0.80 · 0.36
75%measured target: 0.25 · 0.67 · 0.66
100%measured target: 0.69 · 0.65 · 0.38
125% ★measured target: 0.12 · 0.66 · 0.58
Merging Narrator Style __low dials back the dimension: at the recommended 125% dose the measured Narrator Style score moves from a 0.45 baseline to 0.39 (a -0.06 shift). The dose-response is noisy and non-monotonic — the merge is an unreliable knob here. That reach is a steep cost — the biggest hit is to burst-blend (-0.34); across the board genuineness -0.04, burst-blend -0.34, speech quality -0.14. Emotionally it reads as Interest (r 0.53), Longing (0.47) and Contemplation (0.43), and it moves away from Pain, Distress and Fatigue/Exhaustion. On other voice dimensions it co-moves most with Smoothness (0.85), Esthetics (0.83) and Structure (0.81), while pulling down Disfluency, Genuineness and Casual Style. Practical call: merge __low at 100–125%; expect Interest and Smoothness to rise and burst-blend to fall. If you don’t want Interest bleeding in, offset it with a prompt cue toward Pain to keep the narrator style shift without the side-effect.