How two embedding models organize 100,000 voice samples into distinct vocal archetypes
100k audio samples were randomly drawn from 20 shards of
laion/emolia-balanced-5M-subset (Emilia speech corpus).
Each sample was embedded with
VoiceNet/voiceclap-small (768-d, 110M params) and
VoiceNet/voiceclap-large (3584-d, 7B params, Qwen2.5-Omni backbone),
then clustered into 100 groups via GPU K-Means.
The 6 most distinctive clusters per model were selected and captioned by Gemini 3.1 Pro.
Each card shows the 5 samples nearest to the cluster centroid.
768-d · 110M params · 100 clusters · top 6
Emotionally raw, angst-ridden voices conveying deep sorrow, frustration, and tearful distress through a highly dramatic, theatrical delivery.
A fervent, gravelly male voice preaching with intense evangelical zeal, high energy, and a distinct Southern accent.
A gentle, breathy female voice weaves soothing, meditative narratives with a warm, hypnotic, and deeply relaxing cadence.
A high-pitched, bubbly female voice bursting with animated energy, youthful enthusiasm, and exaggerated comedic flair.
A mature, resonant British male voice delivers steady, classic narration with a rich, warm storytelling charm.
An energetic, fast-paced male voice with a slightly nasal, conversational tone perfect for informal podcasts and gaming commentary.
3,584-d · 7B params · 100 clusters · top 6
A wildly energetic, chaotic voice marked by frantic shrieks, breathless laughter, and moments of unhinged comedic panic.
A gentle, conversational female voice characterized by soft, empathetic tones and a warm, slightly breathy texture.
A fiery, high-energy voice blending rhythmic preaching and aggressive spoken-word delivery with a commanding, intense edge.
An earnest, emotionally vulnerable voice sharing intimate personal narratives and supportive advice with gentle, reflective sincerity.
A mature, resonant British male voice delivers steady, reflective narration with a warm, slightly gravelly texture.
An earnest, articulate voice delivers persuasive talks with passionate conviction, engaging clarity, and a slightly breathy texture.