← Overview

Demo 1 — Understanding

Listening along the 57-dimension VoiceNet taxonomy — this is what the agent hears.

Below are clips from EmoLia, each scored live by the four ears. This is the agent’s perception: for any voice it can name the loudest of the 57 dimensions and read its genuineness and blend. We deliberately include high- and low-genuineness cases, natural vs. bolted-on vocal bursts, and clips that are striking on a single dimension (whispered, very bright, rough, high-arousal, nasal…) so you can hear what each number means. It is what makes the generation loop possible — you cannot optimise what you cannot measure.

Very genuine — sounds spontaneous, unrehearsed
EN genuineness 5.1 · blend 5.2
Casual Style 5.1Perceived Gender 4.3Conversational Style 4.3
Very genuine — sounds spontaneous, unrehearsed
EN genuineness 3.9 · blend 3.2
Casual Style 5.7Emphasis 5.1Arousal 4.9
Low genuineness — reads as rehearsed / recited
EN genuineness 0.1 · blend 0.0
Formal Style 3.7Monologue Style 3.7Newsreader Style 3.6
Low genuineness — reads as rehearsed / recited
EN genuineness 0.0 · blend 1.6
Narrator Style 4.3Monologue Style 3.8Emphasis 3.4
Natural vocal burst — blends into the speech
EN genuineness 1.3 · blend 6.5
Fullness 5.6Whisper-Talk Style 5.4Warmth 5.1
A burst that does not blend well
EN genuineness 1.2 · blend 0.0
Dramatic Style 5.8Perceived Gender 5.4Casual Style 5.3
Striking: whispered delivery (Whisper-Talk Style 6.0)
FR genuineness 1.4 · blend 0.9
Whisper-Talk Style 6.0Teacher/Didactic Style 4.7ASMR Style 4.4
Striking: very bright timbre (Brightness 4.2)
EN genuineness 0.9 · blend 1.1
Cartoonish Style 5.4Storytelling Style 5.4Register 5.2
Striking: rough / gravelly voice (Roughness 4.6)
FR genuineness 1.1 · blend 1.5
Fullness 5.5Chest Resonance 5.1Storytelling Style 5.0
Striking: high arousal / intensity (Arousal 6.0)
KO genuineness 0.8 · blend 0.8
Arousal 6.0Attack 6.0Dynamic Arc 6.0
Striking: fast tempo (Tempo 4.9)
FR genuineness 1.8 · blend 5.3
Stance 5.3Dramatic Style 4.9Tempo 4.9
Striking: nasal resonance (Nasal Resonance 5.1)
EN genuineness 1.9 · blend 4.5
Nasal Resonance 5.1Dramatic Style 5.0Storytelling Style 5.0
Striking: vulnerable delivery (Vulnerability 5.0)
FR genuineness 2.9 · blend 6.3
Whisper-Talk Style 5.8Vulnerability 5.0Casual Style 4.6
Striking: warm timbre (Warmth 5.3)
FR genuineness 0.7 · blend 3.2
Fullness 5.8Warmth 5.3Narrator Style 5.1
Striking: deep chest resonance (Chest Resonance 4.9)
JA genuineness 1.2 · blend 3.2
Perceived Gender 5.4Fullness 5.2Chest Resonance 4.9
Anonymized companion for double-blind review. All voices are synthetic, generated by an open-weight text-to-speech model with no reference recordings; all scores are produced by the perceptual models described here. No author, institution, or repository identifiers are included by design.