← all dimensions · markdown (agent skill) · manual home
Respiration VoiceNet dimension guide
Physical Productioncode vn_RESPslot 57
This dimension measures the audibility and laboriousness of breathing. 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.
Respiration controls the audibility and laboriousness of breathing. The __high LoRA shifts it +0.16 at 100% (worst cost: speech quality -0.29); the __low LoRA shifts it -0.24 at 125% (worst cost: burst-blend -0.36). The __low direction is the stronger, more reliable knob here.
▲ HIGH — Respiration __high
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.442 | — | — | — | — |
| 25% | 0.497 | +0.05 | -0.02 | -0.10 | -0.07 |
| 50% | 0.460 | +0.02 | -0.01 | -0.12 | -0.06 |
| 75% | 0.533 | +0.09 | +0.04 | -0.13 | -0.15 |
| 100% ★ | 0.598 | +0.16 | +0.05 | -0.23 | -0.29 |
| 125% | 0.533 | +0.09 | -0.01 | -0.25 | -0.30 |
EmoNet emotion correlation
Amusement+0.49
Fatigue/Exhaustion+0.39
Teasing+0.38
Contemplation-0.69
Affection-0.44
Contentment-0.42
VoiceNet dimension correlation
Casual Style+0.71
Volatility+0.67
Disfluency+0.63
Structure-0.66
Recording Quality-0.62
Formal Style-0.61
Listen — 3 samples per dose
25%measured target: 0.42 · 0.58 · 0.46
50%measured target: 0.73 · 0.43 · 0.50
75%measured target: 0.61 · 0.40 · 0.54
100% ★measured target: 0.45 · 0.62 · 0.63
125%measured target: 0.67 · 0.49 · 0.32
Merging Respiration __high pushes the dimension: at the recommended 100% dose the measured Respiration score moves from a 0.44 baseline to 0.60 (a +0.16 shift). The dose-response is mostly ordered but noisy between doses. That reach is a steep cost — the biggest hit is to speech quality (-0.29); across the board genuineness +0.05, burst-blend -0.23, speech quality -0.29. Emotionally it reads as Amusement (r 0.49), Fatigue/Exhaustion (0.39) and Teasing (0.38), and it moves away from Contemplation, Affection and Contentment. On other voice dimensions it co-moves most with Casual Style (0.71), Volatility (0.67) and Disfluency (0.63), while pulling down Structure, Recording Quality and Formal Style. Practical call: merge __high at 75–100%; expect Amusement and Casual Style to rise and speech quality to fall. If you don’t want Amusement bleeding in, offset it with a prompt cue toward Contemplation to keep the respiration shift without the side-effect.
▼ LOW — Respiration __low
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.442 | — | — | — | — |
| 25% | 0.374 | -0.07 | -0.05 | -0.03 | +0.05 |
| 50% | 0.337 | -0.11 | -0.04 | -0.19 | +0.06 |
| 75% | 0.366 | -0.08 | -0.10 | -0.24 | +0.05 |
| 100% | 0.355 | -0.09 | -0.12 | -0.30 | +0.12 |
| 125% ★ | 0.207 | -0.24 | -0.12 | -0.36 | +0.08 |
EmoNet emotion correlation
Fatigue/Exhaustion+0.39
Anger+0.39
Fear+0.35
Emotional Numbness-0.45
Concentration-0.20
Contemplation-0.19
VoiceNet dimension correlation
Genuineness+0.62
Disfluency+0.60
Volatility+0.60
Newsreader Style-0.65
Formal Style-0.58
Structure-0.57
Listen — 3 samples per dose
25%measured target: 0.40 · 0.40 · 0.22
50%measured target: 0.19 · 0.19 · 0.26
75%measured target: 0.29 · 0.57 · 0.36
100%measured target: 0.29 · 0.31 · 0.66
125% ★measured target: 0.16 · 0.24 · 0.25
Merging Respiration __low damps the dimension: at the recommended 125% dose the measured Respiration score moves from a 0.44 baseline to 0.21 (a -0.24 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 (-0.36); across the board genuineness -0.12, burst-blend -0.36, speech quality +0.08. Emotionally it reads as Fatigue/Exhaustion (r 0.39), Anger (0.39) and Fear (0.35), and it moves away from Emotional Numbness, Concentration and Contemplation. On other voice dimensions it co-moves most with Genuineness (0.62), Disfluency (0.60) and Volatility (0.60), while pulling down Newsreader Style, Formal Style and Structure. Practical call: merge __low at 100–125%; expect Fatigue/Exhaustion and Genuineness to rise and burst-blend to fall. If you don’t want Fatigue/Exhaustion bleeding in, offset it with a prompt cue toward Emotional Numbness to keep the respiration shift without the side-effect.