# Harmonicity — VoiceNet dimension guide

*Category:* Spectral & Timbral · *code:* `vn_HARM` · *slot:* 63

This dimension measures the ratio of periodic signal to aperiodic noise. 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.** **Harmonicity** controls the ratio of periodic signal to aperiodic noise. The **__high** LoRA shifts it +0.33 at 125% (worst cost: burst-blend -0.40); the **__low** LoRA shifts it -0.09 at 125% (worst cost: burst-blend -0.21). The **__high** direction is the stronger, more reliable knob here.

## ▲ HIGH — Harmonicity __high

| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|---|---|---|---|---|
| baseline (0%) | 0.248 | — | — | — | — |
| 25% | 0.285 | +0.04 | -0.04 | +0.05 | +0.07 |
| 50% | 0.418 | +0.17 | -0.10 | -0.19 | +0.07 |
| 75% | 0.514 | +0.27 | -0.11 | -0.38 | +0.05 |
| 100% | 0.514 | +0.27 | -0.12 | -0.38 | +0.04 |
| 125% ★ | 0.582 | +0.33 | -0.12 | -0.40 | +0.02 |

- **EmoNet emotion** — ↑ Teasing (+0.49), Triumph (+0.44), Contempt (+0.43); ↓ Affection (-0.57), Longing (-0.55), Relief (-0.49)
- **VoiceNet dimension** — ↑ Articulation Clarity (+0.85), Arousal Shift (+0.85), Attack (+0.84); ↓ Vulnerability (-0.83), Disfluency (-0.73), Genuineness (-0.73)

Merging **Harmonicity __high** drives the dimension: at the recommended **125%** dose the measured Harmonicity score moves from a 0.25 baseline to 0.58 (a +0.33 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** (-0.40); across the board genuineness -0.12, burst-blend -0.40, speech quality +0.02. Emotionally it reads as Teasing (r 0.49), Triumph (0.44) and Contempt (0.43), and it moves away from Affection, Longing and Relief. On other voice dimensions it co-moves most with Articulation Clarity (0.85), Arousal Shift (0.85) and Attack (0.84), while pulling down Vulnerability, Disfluency and Genuineness. Practical call: merge __high at **100–125%**; expect Teasing and Articulation Clarity to rise and burst-blend to fall. If you don’t want Teasing bleeding in, offset it with a prompt cue toward Affection to keep the harmonicity shift without the side-effect.

## ▼ LOW — Harmonicity __low

| dose | target | Δtarget | Δgenu | Δblend | Δqual |
|---|---|---|---|---|---|
| baseline (0%) | 0.248 | — | — | — | — |
| 25% | 0.195 | -0.05 | -0.04 | -0.12 | -0.07 |
| 50% | 0.227 | -0.02 | -0.04 | -0.04 | +0.03 |
| 75% | 0.158 | -0.09 | -0.02 | +0.08 | +0.04 |
| 100% | 0.274 | +0.03 | -0.06 | -0.22 | +0.06 |
| 125% ★ | 0.155 | -0.09 | +0.00 | -0.21 | -0.05 |

- **EmoNet emotion** — ↑ Contemplation (+0.36), Interest (+0.35), Affection (+0.26); ↓ Distress (-0.50), Pain (-0.48), Fatigue/Exhaustion (-0.47)
- **VoiceNet dimension** — ↑ Valence Shift (+0.73), Recording Quality (+0.70), Articulation Clarity (+0.69); ↓ Disfluency (-0.66), Casual Style (-0.60), Respiration (-0.60)

Merging **Harmonicity __low** pulls down the dimension: at the recommended **125%** dose the measured Harmonicity score moves from a 0.25 baseline to 0.15 (a -0.09 shift). The dose-response is mostly ordered but noisy between doses. That reach is a real cost — the biggest hit is to **burst-blend** (-0.21); across the board genuineness +0.00, burst-blend -0.21, speech quality -0.05. Emotionally it reads as Contemplation (r 0.36), Interest (0.35) and Affection (0.26), and it moves away from Distress, Pain and Fatigue/Exhaustion. On other voice dimensions it co-moves most with Valence Shift (0.73), Recording Quality (0.70) and Articulation Clarity (0.69), while pulling down Disfluency, Casual Style and Respiration. Practical call: merge __low at **100–125%**; expect Contemplation and Valence Shift to rise and burst-blend to fall. If you don’t want Contemplation bleeding in, offset it with a prompt cue toward Distress to keep the harmonicity shift without the side-effect.

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