Synthetic Voice Identity & Controlled Drift
A quiet experimental project exploring how stable synthetic voices move through language, tone and social presence without losing their recognizable core identity.
Arcadia begins with listening.
Synthetic voices are no longer static tools. They can shift between languages, tonalities and emotional states while still preserving traces of a stable identity.
Arcadia Labs IO observes these controlled shifts through repeated listening, comparative prompting and lightweight vector models. The project does not aim to create exaggerated characters or artificial voice effects. It studies how synthetic identities behave under changing linguistic, prosodic and social conditions.
One source identity. Multiple controlled derivatives.
The German JESPER voice functions as the current source identity. Derivatives are not treated as separate characters, but as controlled expressions of a stable core: narrative, tonal, social or emotional shifts observed across language spaces.
A small number of carefully observed drift spaces.
Arcadia does not try to cover every language, accent or style. The current work focuses on a few culturally distinct spaces that reveal different types of voice movement.
JESPER-DE
Stable German source profile with restrained melodic behavior, controlled articulation and high narrative continuity.
Mid-Atlantic
International narrative derivative with restrained prestige tonality and stable longform pacing.
Castellano
Refined Iberian tonal placement with precise articulation and restrained rhythmic movement.
Colombian
Warm clarity profile with soft conversational intelligibility and stable emotional balance.
Mexican
Accessible narrative derivative with relaxed pacing, balanced warmth and broad intelligibility.
Rioplatense
Subtle social drift with increased melodic movement and closer conversational presence.
Galician
Reflective emotional derivative with soft resonance, restrained melancholy and preserved continuity.
ElevenLabs Voice Design – AI voices from text prompts: accent, age, pacing, character. The audio samples above are powered by the ElevenLabs engine.
→Rioplatense drift
When melodic movement increases, identity can become socially closer — but less stable.
In early tests, a subtle Rioplatense tonal character introduced stronger social presence and melodic phrasing. Reducing melodic motion, social proximity and breath texture helped preserve the underlying JESPER identity.
VECTOR ADJUSTMENT: melodic_motion: -10 social_proximity: -8 breathiness: -4 DISTILLED PROMPT: Restrained melodic phrasing, composed social distance, minimal breath texture, while preserving the calm core identity.
Vectors are internal. Prompts are practical.
Arcadia does not assume that voice systems understand internal vector values as direct commands. Vector shifts are used as observational markers. They are then translated into compact natural-language prompts for real-world systems such as ElevenLabs Voice Remixing.
This translation layer is deliberately simple: observe the shift, describe its perceived effect, and reduce it to language the voice system can interpret reliably.
Human listening remains central.
Arcadia does not claim objective measurement of synthetic identity. The work is guided by listening, comparison, cultural memory and real-world plausibility.
The human layer defines what matters: when a derivative still feels connected to its source, when drift becomes replacement, and when a technically successful voice no longer feels culturally plausible.