← all dimensions · markdown (agent skill) · manual home
Recording Quality VoiceNet dimension guide
Language & Recordingcode vn_RCQLslot 72
This dimension measures the technical fidelity and cleanliness of the recording. 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.
Recording Quality controls the technical fidelity and cleanliness of the recording. The __high LoRA shifts it +0.14 at 125% (worst cost: burst-blend -0.16); the __low LoRA shifts it -0.30 at 125% (worst cost: speech quality -0.31). The __low direction is the stronger, more reliable knob here.
▲ HIGH — Recording Quality __high
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.518 | — | — | — | — |
| 25% | 0.622 | +0.10 | -0.04 | +0.03 | +0.11 |
| 50% | 0.634 | +0.12 | -0.06 | -0.06 | +0.12 |
| 75% | 0.638 | +0.12 | -0.06 | -0.07 | +0.10 |
| 100% | 0.648 | +0.13 | -0.08 | -0.20 | +0.13 |
| 125% ★ | 0.654 | +0.14 | -0.11 | -0.16 | +0.15 |
EmoNet emotion correlation
Emotional Numbness+0.52
Contemplation+0.47
Contentment+0.41
Fatigue/Exhaustion-0.57
Impatience and Irritability-0.56
Longing-0.48
VoiceNet dimension correlation
Teacher/Didactic Style+0.84
Esthetics+0.83
Valence+0.79
Disfluency-0.77
Cognitive Load-0.69
Respiration-0.58
Listen — 3 samples per dose
25%measured target: 0.60 · 0.77 · 0.68
50%measured target: 0.78 · 0.68 · 0.75
75%measured target: 0.68 · 0.69 · 0.59
100%measured target: 0.71 · 0.64 · 0.65
125% ★measured target: 0.69 · 0.71 · 0.84
Merging Recording Quality __high ramps the dimension: at the recommended 125% dose the measured Recording Quality score moves from a 0.52 baseline to 0.65 (a +0.14 shift). The dose-response is clean and monotonic — each step pushes the dimension further. That reach is a real cost — the biggest hit is to burst-blend (-0.16); across the board genuineness -0.11, burst-blend -0.16, speech quality +0.15. Emotionally it reads as Emotional Numbness (r 0.52), Contemplation (0.47) and Contentment (0.41), and it moves away from Fatigue/Exhaustion, Impatience and Irritability and Longing. On other voice dimensions it co-moves most with Teacher/Didactic Style (0.84), Esthetics (0.83) and Valence (0.79), while pulling down Disfluency, Cognitive Load and Respiration. Practical call: merge __high at 100–125%; expect Emotional Numbness and Teacher/Didactic Style to rise and burst-blend to fall. If you don’t want Emotional Numbness bleeding in, offset it with a prompt cue toward Fatigue/Exhaustion to keep the recording quality shift without the side-effect.
▼ LOW — Recording Quality __low
Dose-response
| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|
| baseline (0%) | 0.518 | — | — | — | — |
| 25% | 0.452 | -0.07 | +0.01 | -0.05 | -0.07 |
| 50% | 0.345 | -0.17 | +0.06 | +0.01 | -0.17 |
| 75% | 0.433 | -0.08 | +0.02 | -0.14 | -0.04 |
| 100% | 0.271 | -0.25 | +0.04 | -0.15 | -0.26 |
| 125% ★ | 0.216 | -0.30 | +0.04 | -0.22 | -0.31 |
EmoNet emotion correlation
Contemplation+0.80
Interest+0.77
Affection+0.73
Fatigue/Exhaustion-0.71
Pain-0.57
Distress-0.46
VoiceNet dimension correlation
Esthetics+0.94
Smoothness+0.90
Warmth+0.90
Disfluency-0.78
Playful Style-0.76
Genuineness-0.57
Listen — 3 samples per dose
25%measured target: 0.56 · 0.56 · 0.56
50%measured target: 0.17 · 0.17 · 0.36
75%measured target: 0.52 · 0.20 · 0.35
100%measured target: 0.24 · 0.15 · 0.18
125% ★measured target: 0.41 · 0.13 · 0.22
Merging Recording Quality __low suppresses the dimension: at the recommended 125% dose the measured Recording Quality score moves from a 0.52 baseline to 0.22 (a -0.30 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 (-0.31); across the board genuineness +0.04, burst-blend -0.22, speech quality -0.31. Emotionally it reads as Contemplation (r 0.80), Interest (0.77) and Affection (0.73), and it moves away from Fatigue/Exhaustion, Pain and Distress. On other voice dimensions it co-moves most with Esthetics (0.94), Smoothness (0.90) and Warmth (0.90), while pulling down Disfluency, Playful Style and Genuineness. Practical call: merge __low at 100–125%; expect Contemplation and Esthetics to rise and speech quality to fall. If you don’t want Contemplation bleeding in, offset it with a prompt cue toward Fatigue/Exhaustion to keep the recording quality shift without the side-effect.