← all dimensions · markdown (agent skill) · manual home
Conversational Style VoiceNet dimension guide
Speaking Stylecode vn_S_CONVslot 83
This dimension measures how conversational and spontaneous the delivery sounds. 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.
Conversational Style controls how conversational and spontaneous the delivery sounds. The __high LoRA shifts it +0.12 at 125% (worst cost: burst-blend -0.15); the __low LoRA shifts it -0.03 at 125% (worst cost: burst-blend -0.32). The __high direction is the stronger, more reliable knob here.
▲ HIGH — Conversational Style __high
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.046 | — | — | — | — |
| 25% | 0.038 | -0.01 | -0.07 | -0.11 | +0.05 |
| 50% | 0.043 | +0.00 | -0.03 | -0.01 | +0.03 |
| 75% | 0.094 | +0.05 | -0.07 | -0.20 | +0.08 |
| 100% | 0.143 | +0.10 | -0.08 | -0.20 | +0.05 |
| 125% ★ | 0.166 | +0.12 | -0.08 | -0.15 | +0.08 |
EmoNet emotion correlation
Amusement+0.58
Embarrassment+0.49
Teasing+0.43
Longing-0.45
Contemplation-0.36
Interest-0.35
VoiceNet dimension correlation
Dynamic Arc+0.74
Arousal Shift+0.62
Playful Style+0.62
Vocal-burst blend-0.40
Cognitive Load-0.40
Whisper-Talk Style-0.25
Listen — 3 samples per dose
25%measured target: 0.00 · 0.01 · 0.00
50%measured target: 0.00 · 0.04 · 0.00
75%measured target: 0.44 · 0.03 · 0.02
100%measured target: 0.27 · 0.43 · 0.03
125% ★measured target: 0.53 · 0.03 · 0.07
Merging Conversational Style __high raises the dimension: at the recommended 125% dose the measured Conversational Style score moves from a 0.05 baseline to 0.17 (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 (-0.15); across the board genuineness -0.08, burst-blend -0.15, speech quality +0.08. Emotionally it reads as Amusement (r 0.58), Embarrassment (0.49) and Teasing (0.43), and it moves away from Longing, Contemplation and Interest. On other voice dimensions it co-moves most with Dynamic Arc (0.74), Arousal Shift (0.62) and Playful Style (0.62), while pulling down Vocal-burst blend, Cognitive Load and Whisper-Talk Style. Practical call: merge __high at 100–125%; expect Amusement and Dynamic Arc to rise and burst-blend to fall. If you don’t want Amusement bleeding in, offset it with a prompt cue toward Longing to keep the conversational style shift without the side-effect.
▼ LOW — Conversational Style __low
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.046 | — | — | — | — |
| 25% | 0.021 | -0.03 | +0.02 | +0.16 | +0.00 |
| 50% | 0.044 | +0.00 | -0.04 | -0.07 | +0.04 |
| 75% | 0.021 | -0.02 | -0.05 | -0.15 | +0.03 |
| 100% | 0.124 | +0.08 | +0.01 | -0.17 | +0.00 |
| 125% ★ | 0.013 | -0.03 | -0.06 | -0.32 | -0.04 |
EmoNet emotion correlation
Concentration+0.17
Embarrassment+0.16
Emotional Numbness+0.14
Interest-0.41
Longing-0.38
Affection-0.37
VoiceNet dimension correlation
Playful Style+0.55
Casual Style+0.52
Content Appropriateness (3-point Scale)+0.51
Monologue Style-0.52
Narrator Style-0.50
Formal Style-0.46
Listen — 3 samples per dose
25%measured target: 0.00 · 0.00 · 0.00
50%measured target: 0.05 · 0.00 · 0.00
75%measured target: 0.11 · 0.00 · 0.00
100%measured target: 0.00 · 0.23 · 0.00
125% ★measured target: 0.00 · 0.00 · 0.00
Merging Conversational Style __low pulls down the dimension: at the recommended 125% dose the measured Conversational Style score moves from a 0.05 baseline to 0.01 (a -0.03 shift). The dose-response is mostly ordered but noisy between doses. That reach is a steep cost — the biggest hit is to burst-blend (-0.32); across the board genuineness -0.06, burst-blend -0.32, speech quality -0.04. Emotionally it reads as Concentration (r 0.17), Embarrassment (0.16) and Emotional Numbness (0.14), and it moves away from Interest, Longing and Affection. On other voice dimensions it co-moves most with Playful Style (0.55), Casual Style (0.52) and Content Appropriateness (3-point Scale) (0.51), while pulling down Monologue Style, Narrator Style and Formal Style. Practical call: merge __low at 100–125%; expect Concentration and Playful Style to rise and burst-blend to fall. If you don’t want Concentration bleeding in, offset it with a prompt cue toward Interest to keep the conversational style shift without the side-effect.