← all dimensions · markdown (agent skill) · manual home
Emphasis VoiceNet dimension guide
Rhythm & Timingcode vn_EMPHslot 45
This dimension measures how strongly the speaker stresses words to build an informational hierarchy. 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.
Emphasis controls how strongly the speaker stresses words to build an informational hierarchy. The __high LoRA shifts it +0.59 at 100% (worst cost: burst-blend -0.30); the __low LoRA shifts it -0.08 at 100% (worst cost: burst-blend -0.15). The __high direction is the stronger, more reliable knob here.
▲ HIGH — Emphasis __high
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.367 | — | — | — | — |
| 25% | 0.458 | +0.09 | -0.01 | -0.12 | +0.02 |
| 50% | 0.624 | +0.26 | -0.06 | -0.30 | +0.00 |
| 75% | 0.856 | +0.49 | -0.10 | -0.39 | -0.08 |
| 100% ★ | 0.952 | +0.59 | -0.08 | -0.30 | -0.19 |
| 125% | 0.938 | +0.57 | -0.08 | -0.34 | -0.23 |
EmoNet emotion correlation
Triumph+0.71
Elation+0.67
Anger+0.66
Contemplation-0.79
Relief-0.65
Contentment-0.60
VoiceNet dimension correlation
Stance+0.97
Arousal+0.97
Dynamic Arc+0.96
Vulnerability-0.92
Warmth-0.90
Whisper-Talk Style-0.85
Listen — 3 samples per dose
25%measured target: 0.58 · 0.36 · 0.30
50%measured target: 0.54 · 0.42 · 0.72
75%measured target: 0.79 · 0.92 · 0.81
100% ★measured target: 1.00 · 0.96 · 0.85
125%measured target: 0.85 · 0.98 · 0.95
Merging Emphasis __high raises the dimension: at the recommended 100% dose the measured Emphasis score moves from a 0.37 baseline to 0.95 (a +0.59 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 (-0.30); across the board genuineness -0.08, burst-blend -0.30, speech quality -0.19. Emotionally it reads as Triumph (r 0.71), Elation (0.67) and Anger (0.66), and it moves away from Contemplation, Relief and Contentment. On other voice dimensions it co-moves most with Stance (0.97), Arousal (0.97) and Dynamic Arc (0.96), while pulling down Vulnerability, Warmth and Whisper-Talk Style. Practical call: merge __high at 75–100%; expect Triumph and Stance 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 emphasis shift without the side-effect.
▼ LOW — Emphasis __low
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.367 | — | — | — | — |
| 25% | 0.311 | -0.06 | -0.01 | -0.23 | -0.18 |
| 50% | 0.389 | +0.02 | -0.03 | -0.15 | -0.03 |
| 75% | 0.363 | +0.00 | -0.04 | -0.18 | -0.01 |
| 100% ★ | 0.287 | -0.08 | -0.03 | -0.15 | -0.01 |
| 125% | 0.287 | -0.08 | -0.02 | -0.04 | +0.04 |
EmoNet emotion correlation
Interest+0.63
Malevolence/Malice+0.44
Hope/Enthusiasm/Optimism+0.41
Fatigue/Exhaustion-0.48
Pain-0.42
Distress-0.37
VoiceNet dimension correlation
Focus+0.74
Tempo+0.70
Storytelling Style+0.69
Vulnerability-0.49
Conversational Style-0.46
Disfluency-0.43
Listen — 3 samples per dose
25%measured target: 0.48 · 0.33 · 0.11
50%measured target: 0.16 · 0.44 · 0.43
75%measured target: 0.24 · 0.41 · 0.47
100% ★measured target: 0.27 · 0.41 · 0.44
125%measured target: 0.23 · 0.33 · 0.44
Merging Emphasis __low suppresses the dimension: at the recommended 100% dose the measured Emphasis score moves from a 0.37 baseline to 0.29 (a -0.08 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 (-0.15); across the board genuineness -0.03, burst-blend -0.15, speech quality -0.01. Emotionally it reads as Interest (r 0.63), Malevolence/Malice (0.44) and Hope/Enthusiasm/Optimism (0.41), and it moves away from Fatigue/Exhaustion, Pain and Distress. On other voice dimensions it co-moves most with Focus (0.74), Tempo (0.70) and Storytelling Style (0.69), while pulling down Vulnerability, Conversational Style and Disfluency. Practical call: merge __low at 75–100%; expect Interest and Focus to rise and burst-blend to fall. If you don’t want Interest bleeding in, offset it with a prompt cue toward Fatigue/Exhaustion to keep the emphasis shift without the side-effect.