Working observations from an unfinished field.
Arcadia Notes collects small listening observations, drift behaviors, prompt findings and practical reflections gathered during ongoing synthetic voice experiments.
These are not scientific papers.
The Notes section is intentionally lightweight. It documents practical observations, recurring behaviors and unresolved questions without pretending to offer final theoretical certainty.
Some findings remain unstable. Others may change as voice systems evolve. The value lies in careful listening, comparison and long-term observation.
Small movements often matter more than large ones.
Melodic movement can destabilize identity faster than tonal color.
Early Rioplatense tests revealed that increased melodic motion and social proximity changed the perceived personality of the voice more strongly than accent coloration itself.
Small reductions in melodic movement restored source continuity surprisingly well.
Controlled Iberian placement preserves composure.
Castellano drift remained comparatively stable when tonal precision and restrained rhythm were prioritized over strong regional emphasis.
Excessive regional stylization reduced perceived elegance and source continuity.
Compact natural language often works better than explicit vectors.
Internal vector structures appear useful for orientation, but practical voice systems frequently respond more reliably to concise semantic phrasing.
The strongest results often emerged from reduced prompts rather than parameter-heavy descriptions.
German JESPER behaves more like a source than Mid-Atlantic JESPER.
Continued derivative testing suggests that the German JESPER profile functions as the actual stability baseline, while Mid-Atlantic behaves more like a narrative derivative.
This observation significantly changed the Arcadia source architecture.
Voice systems evolve faster than stable theory.
Arcadia remains intentionally observational.
Voice Remixing systems, multilingual transfer and prompt interpretation models continue to change rapidly. Arcadia therefore avoids rigid claims of technical permanence.
The same prompt may behave differently across model generations and platform updates.
Regional perception changes across audiences and listening contexts.
Smaller prompts sometimes outperform longer technically detailed descriptions.
Final evaluation still depends on listening, memory and contextual plausibility.
The project stays intentionally small.
Arcadia is not attempting to build a universal voice taxonomy or industrial measurement system. The project focuses on a limited number of carefully observed identity movements and practical derivative behaviors.
The emphasis remains on continuity, plausibility and listening quality rather than scale.
Observation requires interpretation.
Arcadia Notes are not autonomous machine conclusions. They are curated observations shaped through repeated listening, comparative evaluation and human interpretation.