← all emotions
Confusion
Best adapter: rank 64 , checkpoint seen 1672 · EmoNet-z +0.54 (baseline 0.302) · blend 0.456 · genu 0.194 · speaker-sim 0.199.
Instruction prompt: GENERAL: A voice that is adult; extremely carrying bewilderment; genuine in delivery.
SCRIPT:
Five neutral sentences, each in English / Deutsch / Français, no reference vs with emotion reference (a same-emotion training clip supplies the voice; the prompt applies the emotion).
Each cell = 5 seeds. We show mean±std of the emotion score and play the highest and lowest seed (top = strongest emotion, bottom = weakest), since a fixed seed pins the sampling and single clips vary.
Sentence 1
no reference
z mean -0.70 ±0.00 · bl 0.2 · gn 0.3 ▲ high -0.70
▼ low -0.70
z mean -0.70 ±0.00 · bl 0.4 · gn 0.1 ▲ high -0.70
▼ low -0.70
z mean -0.28 ±0.51 · bl 0.5 · gn 0.2 ▲ high +0.44
▼ low -0.70
with emo reference
z mean -0.70 ±0.00 · bl 0.3 · gn 0.4 ▲ high -0.70
▼ low -0.70
z mean -0.32 ±0.75 · bl 0.1 · gn 0.2 ▲ high +1.18
▼ low -0.70
z mean -0.27 ±0.86 · bl 0.2 · gn 0.2 ▲ high +1.45
▼ low -0.70
Sentence 2
no reference
z mean -0.15 ±1.10 · bl 0.5 · gn 0.3 ▲ high +2.06
▼ low -0.70
z mean -0.41 ±0.45 · bl 0.5 · gn 0.2 ▲ high +0.49
▼ low -0.70
z mean -0.41 ±0.48 · bl 0.5 · gn 0.2 ▲ high +0.54
▼ low -0.70
with emo reference
z mean -0.67 ±0.06 · bl 0.6 · gn 0.3 ▲ high -0.56
▼ low -0.70
z mean -0.70 ±0.00 · bl 0.4 · gn 0.1 ▲ high -0.69
▼ low -0.70
z mean -0.70 ±0.00 · bl 0.3 · gn 0.3 ▲ high -0.70
▼ low -0.70
Sentence 3
no reference
z mean -0.63 ±0.11 · bl 0.1 · gn 0.2 ▲ high -0.41
▼ low -0.70
z mean +1.93 ±0.39 · bl 0.4 · gn 0.1 ▲ high +2.39
▼ low +1.28
z mean -0.70 ±0.00 · bl 0.5 · gn 0.2 ▲ high -0.70
▼ low -0.70
with emo reference
z mean -0.31 ±0.52 · bl 0.2 · gn 0.3 ▲ high +0.58
▼ low -0.70
z mean +2.03 ±1.26 · bl 0.1 · gn 0.2 ▲ high +3.90
▼ low +0.00
z mean -0.56 ±0.27 · bl 0.3 · gn 0.3 ▲ high -0.03
▼ low -0.70
Sentence 4
no reference
z mean -0.70 ±0.00 · bl 0.1 · gn 0.2 ▲ high -0.70
▼ low -0.70
z mean -0.43 ±0.55 · bl 0.4 · gn 0.1 ▲ high +0.68
▼ low -0.70
z mean -0.70 ±0.00 · bl 0.5 · gn 0.2 ▲ high -0.70
▼ low -0.71
with emo reference
z mean -0.70 ±0.00 · bl 0.2 · gn 0.3 ▲ high -0.70
▼ low -0.70
z mean -0.70 ±0.00 · bl 0.1 · gn 0.2 ▲ high -0.70
▼ low -0.70
z mean -0.70 ±0.00 · bl 0.4 · gn 0.3 ▲ high -0.70
▼ low -0.70
Sentence 5
no reference
z mean -0.57 ±0.27 · bl 0.2 · gn 0.1 ▲ high -0.02
▼ low -0.70
z mean -0.01 ±1.10 · bl 0.6 · gn 0.2 ▲ high +2.16
▼ low -0.70
z mean -0.54 ±0.28 · bl 0.4 · gn 0.2 ▲ high +0.02
▼ low -0.70
with emo reference
z mean -0.28 ±0.84 · bl 0.2 · gn 0.2 ▲ high +1.41
▼ low -0.70
z mean -0.00 ±0.75 · bl 0.1 · gn 0.2 ▲ high +1.04
▼ low -0.68
z mean -0.29 ±0.52 · bl 0.2 · gn 0.1 ▲ high +0.56
▼ low -0.70