The project tests whether vocal identity, articulation, timing, prosody and presence can be treated as separable systems rather than one inseparable finished recording.
The donor supplies intelligible speech, consonant placement and timing. At this point the voice is useful as an articulation source, not as the target identity.
An earlier branch explored self-contained browser synthesis: mathematical excitation, phoneme targets, formant motion, source/filter rendering and a hybrid soft-tract renderer.
Analysis experiments progressively moved from broad empirical voice structure toward speech landmarks and presence, using observed behaviour to derive numerical rules rather than retaining source recordings.
R3 still reshapes a finished donor. The next question is more fundamental: can speech be represented as separate vocal identity/body, articulation/timing and prosody/delivery layers, then recombined without inheriting the donor as a finished voice?
Only deliberately created showcase clips belong here. Private source voices, personal audio and private creative context remain outside the public site. Stage descriptions expose the engineering idea without exposing the material used during private development.