# Valence — VoiceNet dimension guide

*Category:* Emotion & Affect · *code:* `vn_VALN` · *slot:* 54

This dimension measures the global emotional polarity of the performance. Merge the `__high` / `__low` LoRA from `laion/moss-voicenet-dimension-loras` at 25/50/75/100/125%. Costs are shifts vs. the 24-sample neutral baseline.

**TL;DR.** **Valence** controls the global emotional polarity of the performance. The **__high** LoRA shifts it +0.24 at 125% (worst cost: burst-blend -0.29); the **__low** LoRA shifts it -0.01 at 25% (worst cost: burst-blend -0.06). The **__high** direction is the stronger, more reliable knob here.

## ▲ HIGH — Valence __high

| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|---|---|---|---|---|
| baseline (0%) | 0.076 | — | — | — | — |
| 25% | 0.108 | +0.03 | -0.03 | -0.07 | +0.05 |
| 50% | 0.180 | +0.10 | -0.10 | -0.26 | +0.09 |
| 75% | 0.152 | +0.08 | -0.08 | -0.30 | +0.02 |
| 100% | 0.303 | +0.23 | -0.01 | -0.30 | +0.12 |
| 125% ★ | 0.314 | +0.24 | +0.02 | -0.29 | +0.04 |

- **EmoNet emotion** — ↑ Emotional Numbness (+0.44), Concentration (+0.35), Contentment (+0.20); ↓ Interest (-0.56), Fear (-0.48), Malevolence/Malice (-0.43)
- **VoiceNet dimension** — ↑ Nasal Resonance (+0.73), Head Resonance (+0.71), Conversational Style (+0.71); ↓ Perceived Gender (-0.62), Cartoonish Style (-0.59), Dramatic Style (-0.55)

Merging **Valence __high** ramps the dimension: at the recommended **125%** dose the measured Valence score moves from a 0.08 baseline to 0.31 (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** (-0.29); across the board genuineness +0.02, burst-blend -0.29, speech quality +0.04. Emotionally it reads as Emotional Numbness (r 0.44), Concentration (0.35) and Contentment (0.20), and it moves away from Interest, Fear and Malevolence/Malice. On other voice dimensions it co-moves most with Nasal Resonance (0.73), Head Resonance (0.71) and Conversational Style (0.71), while pulling down Perceived Gender, Cartoonish Style and Dramatic Style. Practical call: merge __high at **100–125%**; expect Emotional Numbness and Nasal Resonance to rise and burst-blend to fall. If you don’t want Emotional Numbness bleeding in, offset it with a prompt cue toward Interest to keep the valence shift without the side-effect.

## ▼ LOW — Valence __low

| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|---|---|---|---|---|
| baseline (0%) | 0.076 | — | — | — | — |
| 25% ★ | 0.069 | -0.01 | -0.03 | -0.06 | +0.02 |
| 50% | 0.097 | +0.02 | -0.02 | -0.05 | -0.01 |
| 75% | 0.120 | +0.04 | -0.03 | -0.11 | +0.00 |
| 100% | 0.269 | +0.19 | -0.06 | -0.22 | -0.04 |
| 125% | 0.099 | +0.02 | -0.05 | -0.24 | -0.07 |

- **EmoNet emotion** — ↑ Amusement (+0.39), Teasing (+0.38), Emotional Numbness (+0.19); ↓ Awe (-0.38), Fear (-0.36), Infatuation (-0.36)
- **VoiceNet dimension** — ↑ Head Resonance (+0.78), Brightness (+0.78), Mask Resonance (+0.77); ↓ Cognitive Load (-0.62), Disfluency (-0.50), Perceived Gender (-0.48)

Merging **Valence __low** pulls down the dimension: at the recommended **25%** dose the measured Valence score moves from a 0.08 baseline to 0.07 (a -0.01 shift). The dose-response is noisy and non-monotonic — the merge is an unreliable knob here. That reach is nearly free — the biggest hit is to **burst-blend** (-0.06); across the board genuineness -0.03, burst-blend -0.06, speech quality +0.02. Emotionally it reads as Amusement (r 0.39), Teasing (0.38) and Emotional Numbness (0.19), and it moves away from Awe, Fear and Infatuation. On other voice dimensions it co-moves most with Head Resonance (0.78), Brightness (0.78) and Mask Resonance (0.77), while pulling down Cognitive Load, Disfluency and Perceived Gender. Practical call: merge __low at **a light 25–50%**; expect Amusement and Head Resonance to rise and burst-blend to fall. If you don’t want Amusement bleeding in, offset it with a prompt cue toward Awe to keep the valence shift without the side-effect.

---
[← all dimensions](index.md) · [human/audio version](../../site/dims/VALN.html)