← all dimensions · markdown (agent skill) · manual home
Smoothness VoiceNet dimension guide
Rhythm & Timingcode vn_SMTHslot 42
This dimension measures the regularity of the timing grid and how fluidly units join. 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.
Smoothness controls the regularity of the timing grid and how fluidly units join. The __high LoRA shifts it +0.21 at 125% (worst cost: genuineness -0.09); the __low LoRA shifts it -0.16 at 125% (worst cost: burst-blend -0.27). The __high direction is the stronger, more reliable knob here.
▲ HIGH — Smoothness __high
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.300 | — | — | — | — |
| 25% | 0.324 | +0.02 | -0.04 | -0.04 | +0.01 |
| 50% | 0.361 | +0.06 | -0.04 | -0.07 | +0.09 |
| 75% | 0.401 | +0.10 | -0.06 | +0.01 | +0.11 |
| 100% | 0.444 | +0.14 | -0.11 | -0.06 | +0.14 |
| 125% ★ | 0.505 | +0.21 | -0.09 | -0.08 | +0.13 |
EmoNet emotion correlation
Contemplation+0.41
Concentration+0.35
Sexual Lust+0.28
Fatigue/Exhaustion-0.54
Relief-0.48
Fear-0.42
VoiceNet dimension correlation
Structure+0.86
Narrator Style+0.84
Esthetics+0.83
Disfluency-0.78
Genuineness-0.66
Casual Style-0.65
Listen — 3 samples per dose
25%measured target: 0.49 · 0.32 · 0.35
50%measured target: 0.37 · 0.39 · 0.40
75%measured target: 0.38 · 0.49 · 0.45
100%measured target: 0.41 · 0.59 · 0.31
125% ★measured target: 0.41 · 0.57 · 0.61
Merging Smoothness __high lifts the dimension: at the recommended 125% dose the measured Smoothness score moves from a 0.30 baseline to 0.51 (a +0.21 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 genuineness (-0.09); across the board genuineness -0.09, burst-blend -0.08, speech quality +0.13. Emotionally it reads as Contemplation (r 0.41), Concentration (0.35) and Sexual Lust (0.28), and it moves away from Fatigue/Exhaustion, Relief and Fear. On other voice dimensions it co-moves most with Structure (0.86), Narrator Style (0.84) and Esthetics (0.83), while pulling down Disfluency, Genuineness and Casual Style. Practical call: merge __high at 100–125%; expect Contemplation and Structure to rise and genuineness to fall. If you don’t want Contemplation bleeding in, offset it with a prompt cue toward Fatigue/Exhaustion to keep the smoothness shift without the side-effect.
▼ LOW — Smoothness __low
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.300 | — | — | — | — |
| 25% | 0.237 | -0.06 | +0.02 | -0.13 | -0.10 |
| 50% | 0.248 | -0.05 | -0.04 | -0.17 | -0.05 |
| 75% | 0.272 | -0.03 | +0.01 | -0.10 | -0.11 |
| 100% | 0.261 | -0.04 | -0.03 | -0.32 | -0.15 |
| 125% ★ | 0.140 | -0.16 | -0.03 | -0.27 | -0.26 |
EmoNet emotion correlation
Interest+0.75
Concentration+0.56
Contemplation+0.55
Pain-0.64
Distress-0.63
Impatience and Irritability-0.56
VoiceNet dimension correlation
Esthetics+0.90
Narrator Style+0.89
Recording Quality+0.86
Disfluency-0.77
Playful Style-0.62
Ranting/Angry Style-0.62
Listen — 3 samples per dose
25%measured target: 0.35 · 0.13 · 0.03
50%measured target: 0.00 · 0.30 · 0.37
75%measured target: 0.01 · 0.44 · 0.39
100%measured target: 0.12 · 0.19 · 0.39
125% ★measured target: 0.00 · 0.08 · 0.00
Merging Smoothness __low lowers the dimension: at the recommended 125% dose the measured Smoothness score moves from a 0.30 baseline to 0.14 (a -0.16 shift). The dose-response is mostly ordered but noisy between doses. That reach is a real cost — the biggest hit is to burst-blend (-0.27); across the board genuineness -0.03, burst-blend -0.27, speech quality -0.26. Emotionally it reads as Interest (r 0.75), Concentration (0.56) and Contemplation (0.55), and it moves away from Pain, Distress and Impatience and Irritability. On other voice dimensions it co-moves most with Esthetics (0.90), Narrator Style (0.89) and Recording Quality (0.86), while pulling down Disfluency, Playful Style and Ranting/Angry Style. Practical call: merge __low at 100–125%; expect Interest and Esthetics 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 smoothness shift without the side-effect.