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

Playful Style VoiceNet dimension guide

Speaking Stylecode vn_S_PLAYslot 90
This dimension measures how playful and light 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.

Playful Style controls how playful and light the delivery sounds. The __high LoRA shifts it +0.61 at 125% (worst cost: burst-blend -0.40); the __low LoRA shifts it -0.04 at 75% (worst cost: burst-blend -0.07). The __high direction is the stronger, more reliable knob here.
▲ HIGH — Playful Style __high
Dose-response
dosetargetΔtargetΔgenuΔblendΔqual
baseline (0%)0.108————
25%0.187+0.08-0.01-0.03+0.04
50%0.499+0.39-0.01-0.18-0.08
75%0.534+0.43-0.05-0.34-0.03
100%0.655+0.55-0.02-0.26-0.14
125% ★0.721+0.61-0.05-0.40-0.06
EmoNet emotion correlation
Amusement+0.67
Teasing+0.64
Elation+0.62
Contemplation-0.78
Longing-0.70
Sexual Lust-0.52
VoiceNet dimension correlation
Pitch Range+0.88
Arousal+0.88
Dynamic Arc+0.85
Warmth-0.83
Whisper-Talk Style-0.80
Fullness-0.74
Listen — 3 samples per dose
25%measured target: 0.48 · 0.09 · 0.37
50%measured target: 0.50 · 0.61 · 0.31
75%measured target: 0.11 · 0.43 · 0.69
100%measured target: 0.57 · 0.66 · 0.66
125% ★measured target: 0.68 · 0.58 · 0.77
Merging Playful Style __high ramps the dimension: at the recommended 125% dose the measured Playful Style score moves from a 0.11 baseline to 0.72 (a +0.61 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 (-0.40); across the board genuineness -0.05, burst-blend -0.40, speech quality -0.06. Emotionally it reads as Amusement (r 0.67), Teasing (0.64) and Elation (0.62), and it moves away from Contemplation, Longing and Sexual Lust. On other voice dimensions it co-moves most with Pitch Range (0.88), Arousal (0.88) and Dynamic Arc (0.85), while pulling down Warmth, Whisper-Talk Style and Fullness. Practical call: merge __high at 100–125%; expect Amusement and Pitch Range to rise and burst-blend to fall. If you don’t want Amusement bleeding in, offset it with a prompt cue toward Contemplation to keep the playful style shift without the side-effect.
▼ LOW — Playful Style __low
Dose-response
dosetargetΔtargetΔgenuΔblendΔqual
baseline (0%)0.108————
25%0.181+0.07-0.04-0.11-0.04
50%0.161+0.05-0.07-0.04+0.07
75% ★0.072-0.04-0.06-0.07+0.08
100%0.092-0.02-0.07-0.06+0.12
125%0.087-0.02-0.08-0.06+0.12
EmoNet emotion correlation
Elation+0.36
Doubt+0.34
Amusement+0.34
Contemplation-0.56
Longing-0.28
Concentration-0.26
VoiceNet dimension correlation
Dynamic Arc+0.64
Casual Style+0.53
Volatility+0.42
Warmth-0.81
Whisper-Talk Style-0.71
Fullness-0.68
Listen — 3 samples per dose
25%measured target: 0.10 · 0.39 · 0.04
50%measured target: 0.07 · 0.04 · 0.20
75% ★measured target: 0.09 · 0.08 · 0.00
100%measured target: 0.12 · 0.16 · 0.02
125%measured target: 0.00 · 0.19 · 0.23
Merging Playful Style __low dials back the dimension: at the recommended 75% dose the measured Playful Style score moves from a 0.11 baseline to 0.07 (a -0.04 shift). The dose-response is monotonic but it saturates early, so higher doses add little. That reach is a modest cost — the biggest hit is to burst-blend (-0.07); across the board genuineness -0.06, burst-blend -0.07, speech quality +0.08. Emotionally it reads as Elation (r 0.36), Doubt (0.34) and Amusement (0.34), and it moves away from Contemplation, Longing and Concentration. On other voice dimensions it co-moves most with Dynamic Arc (0.64), Casual Style (0.53) and Volatility (0.42), while pulling down Warmth, Whisper-Talk Style and Fullness. Practical call: merge __low at 50–100%; expect Elation and Dynamic Arc to rise and burst-blend to fall. If you don’t want Elation bleeding in, offset it with a prompt cue toward Contemplation to keep the playful style 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