← all dimensions · markdown (agent skill) · manual home
Warmth VoiceNet dimension guide
Spectral & Timbralcode vn_WARMslot 65
This dimension measures the emphasis on low-mid frequencies (~200–500 Hz). 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.
Warmth controls the emphasis on low-mid frequencies (~200–500 Hz). The __high LoRA shifts it +0.16 at 125% (worst cost: burst-blend -0.13); the __low LoRA shifts it -0.57 at 125% (worst cost: speech quality -0.33). The __low direction is the stronger, more reliable knob here.
▲ HIGH — Warmth __high
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.673 | — | — | — | — |
| 25% | 0.718 | +0.04 | -0.02 | +0.06 | +0.07 |
| 50% | 0.742 | +0.07 | -0.04 | -0.08 | +0.10 |
| 75% | 0.726 | +0.05 | -0.06 | -0.05 | +0.12 |
| 100% | 0.700 | +0.03 | -0.09 | -0.17 | +0.13 |
| 125% ★ | 0.830 | +0.16 | -0.07 | -0.13 | +0.19 |
EmoNet emotion correlation
Contemplation+0.43
Emotional Numbness+0.34
Sexual Lust+0.23
Doubt-0.52
Impatience and Irritability-0.41
Anger-0.37
VoiceNet dimension correlation
Whisper-Talk Style+0.86
Recording Quality+0.74
Monologue Style+0.73
Dynamic Arc-0.70
Ranting/Angry Style-0.69
Playful Style-0.67
Listen — 3 samples per dose
25%measured target: 0.80 · 0.63 · 0.66
50%measured target: 0.73 · 0.83 · 0.73
75%measured target: 0.75 · 0.78 · 0.51
100%measured target: 0.72 · 0.78 · 0.68
125% ★measured target: 0.71 · 0.83 · 0.90
Merging Warmth __high pushes the dimension: at the recommended 125% dose the measured Warmth score moves from a 0.67 baseline to 0.83 (a +0.16 shift). The dose-response is mostly ordered but noisy between doses. That reach is a modest cost — the biggest hit is to burst-blend (-0.13); across the board genuineness -0.07, burst-blend -0.13, speech quality +0.19. Emotionally it reads as Contemplation (r 0.43), Emotional Numbness (0.34) and Sexual Lust (0.23), and it moves away from Doubt, Impatience and Irritability and Anger. On other voice dimensions it co-moves most with Whisper-Talk Style (0.86), Recording Quality (0.74) and Monologue Style (0.73), while pulling down Dynamic Arc, Ranting/Angry Style and Playful Style. Practical call: merge __high at 100–125%; expect Contemplation and Whisper-Talk Style to rise and burst-blend to fall. If you don’t want Contemplation bleeding in, offset it with a prompt cue toward Doubt to keep the warmth shift without the side-effect.
▼ LOW — Warmth __low
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.673 | — | — | — | — |
| 25% | 0.660 | -0.01 | -0.02 | -0.05 | +0.05 |
| 50% | 0.493 | -0.18 | -0.03 | -0.29 | -0.01 |
| 75% | 0.345 | -0.33 | +0.01 | -0.21 | -0.12 |
| 100% | 0.114 | -0.56 | -0.01 | -0.22 | -0.32 |
| 125% ★ | 0.108 | -0.57 | +0.03 | -0.26 | -0.33 |
EmoNet emotion correlation
Contemplation+0.79
Contentment+0.69
Interest+0.63
Astonishment/Surprise-0.64
Pain-0.60
Anger-0.57
VoiceNet dimension correlation
Whisper-Talk Style+0.93
Recording Quality+0.88
Monologue Style+0.88
Arousal-0.86
Playful Style-0.85
Dynamic Arc-0.83
Listen — 3 samples per dose
25%measured target: 0.72 · 0.79 · 0.51
50%measured target: 0.18 · 0.65 · 0.34
75%measured target: 0.10 · 0.11 · 0.19
100%measured target: 0.15 · 0.10 · 0.00
125% ★measured target: 0.20 · 0.11 · 0.05
Merging Warmth __low lowers the dimension: at the recommended 125% dose the measured Warmth score moves from a 0.67 baseline to 0.11 (a -0.57 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 speech quality (-0.33); across the board genuineness +0.03, burst-blend -0.26, speech quality -0.33. Emotionally it reads as Contemplation (r 0.79), Contentment (0.69) and Interest (0.63), and it moves away from Astonishment/Surprise, Pain and Anger. On other voice dimensions it co-moves most with Whisper-Talk Style (0.93), Recording Quality (0.88) and Monologue Style (0.88), while pulling down Arousal, Playful Style and Dynamic Arc. Practical call: merge __low at 100–125%; expect Contemplation and Whisper-Talk Style to rise and speech quality to fall. If you don’t want Contemplation bleeding in, offset it with a prompt cue toward Astonishment/Surprise to keep the warmth shift without the side-effect.