← all dimensions · markdown (agent skill) · manual home
Whisper-Talk Style VoiceNet dimension guide
Speaking Stylecode vn_S_WHISslot 93
This dimension measures how much the voice drops into whisper-talk. 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.
Whisper-Talk Style controls how much the voice drops into whisper-talk. The __high LoRA shifts it +0.27 at 100% (worst cost: burst-blend -0.13); the __low LoRA shifts it -0.45 at 100% (worst cost: burst-blend -0.34). The __low direction is the stronger, more reliable knob here.
▲ HIGH — Whisper-Talk Style __high
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.557 | — | — | — | — |
| 25% | 0.540 | -0.02 | -0.02 | -0.02 | +0.05 |
| 50% | 0.728 | +0.17 | -0.06 | -0.03 | +0.10 |
| 75% | 0.798 | +0.24 | -0.09 | -0.06 | +0.17 |
| 100% ★ | 0.830 | +0.27 | -0.09 | -0.13 | +0.16 |
| 125% | 0.803 | +0.25 | -0.10 | -0.08 | +0.21 |
EmoNet emotion correlation
Emotional Numbness+0.53
Contemplation+0.43
Concentration+0.43
Relief-0.59
Confusion-0.46
Fatigue/Exhaustion-0.41
VoiceNet dimension correlation
Monologue Style+0.85
Warmth+0.84
Esthetics+0.83
Casual Style-0.76
Playful Style-0.71
Disfluency-0.71
Listen — 3 samples per dose
25%measured target: 0.13 · 0.49 · 0.70
50%measured target: 0.67 · 0.73 · 0.91
75%measured target: 0.84 · 0.80 · 0.81
100% ★measured target: 0.79 · 0.83 · 0.78
125%measured target: 0.74 · 0.98 · 0.78
Merging Whisper-Talk Style __high raises the dimension: at the recommended 100% dose the measured Whisper-Talk Style score moves from a 0.56 baseline to 0.83 (a +0.27 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 (-0.13); across the board genuineness -0.09, burst-blend -0.13, speech quality +0.16. Emotionally it reads as Emotional Numbness (r 0.53), Contemplation (0.43) and Concentration (0.43), and it moves away from Relief, Confusion and Fatigue/Exhaustion. On other voice dimensions it co-moves most with Monologue Style (0.85), Warmth (0.84) and Esthetics (0.83), while pulling down Casual Style, Playful Style and Disfluency. Practical call: merge __high at 75–100%; expect Emotional Numbness and Monologue Style to rise and burst-blend to fall. If you don’t want Emotional Numbness bleeding in, offset it with a prompt cue toward Relief to keep the whisper-talk style shift without the side-effect.
▼ LOW — Whisper-Talk Style __low
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.557 | — | — | — | — |
| 25% | 0.467 | -0.09 | +0.01 | -0.08 | -0.02 |
| 50% | 0.446 | -0.11 | -0.01 | -0.12 | -0.03 |
| 75% | 0.230 | -0.33 | -0.08 | -0.39 | -0.02 |
| 100% ★ | 0.106 | -0.45 | -0.07 | -0.34 | -0.06 |
| 125% | 0.121 | -0.44 | -0.05 | -0.24 | -0.04 |
EmoNet emotion correlation
Contemplation+0.69
Awe+0.49
Concentration+0.44
Triumph-0.53
Amusement-0.52
Impatience and Irritability-0.47
VoiceNet dimension correlation
Warmth+0.90
Monologue Style+0.83
Voice Age+0.73
Playful Style-0.75
Tension-0.73
Dynamic Arc-0.71
Listen — 3 samples per dose
25%measured target: 0.36 · 0.48 · 0.26
50%measured target: 0.52 · 0.32 · 0.20
75%measured target: 0.38 · 0.21 · 0.00
100% ★measured target: 0.42 · 0.25 · 0.00
125%measured target: 0.06 · 0.24 · 0.05
Merging Whisper-Talk Style __low pulls down the dimension: at the recommended 100% dose the measured Whisper-Talk Style score moves from a 0.56 baseline to 0.11 (a -0.45 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 (-0.34); across the board genuineness -0.07, burst-blend -0.34, speech quality -0.06. Emotionally it reads as Contemplation (r 0.69), Awe (0.49) and Concentration (0.44), and it moves away from Triumph, Amusement and Impatience and Irritability. On other voice dimensions it co-moves most with Warmth (0.90), Monologue Style (0.83) and Voice Age (0.73), while pulling down Playful Style, Tension and Dynamic Arc. Practical call: merge __low at 75–100%; expect Contemplation and Warmth to rise and burst-blend to fall. If you don’t want Contemplation bleeding in, offset it with a prompt cue toward Triumph to keep the whisper-talk style shift without the side-effect.