N'Ko Synthesis Overview

N'Ko Synthesis Overview

N'Ko synthesis is where movement, speech, inscription, and cultural memory meet.
It is not a claim that the movement stack and N'Ko ASR already share one trained
model.

Core Claim

Both sides solve a related problem:

continuous culturally grounded signal
  -> temporally informed state
  -> authority / correction / safety
  -> output that preserves source specificity

For N'Ko speech, the signal is audio. For computational choreography, the signal
is body movement. The systems can share architectural lessons without sharing
weights or pretending to be the same implementation.

N'Ko ASR

Verified anchor:

• trajectory-biased Transformer CTC;
• local N'Ko ASR code in `Desktop/nko-brain-scanner`;
• trajectory bias through audio trajectory scalars and attention-bias layers.

MAOE status:

- MAOE-style routing/correction exists around ASR output;
- it is not the acoustic CTC anchor;
- it still needs same-snapshot replay/evaluation before being called an ASR
improvement.

Motion Inscription

Motion inscription is already implemented as a controlled claim lane:

MotionMixApp / LUME CC
  -> EchelonBridge.step()
  -> Echelon dynamics
  -> ClaimBridgeService
  -> Rust ClaimBridge
  -> inscription mutation

This is not natural sentence generation from arbitrary gesture. It is structured
motion claim detection with provenance.

Movement Stack

Use concrete current terms:

• EchelonCore;
• LatentUpdater;
• SANPipeline;
• DiffusionService;
• BodyTruth;
• AirDeck;
• ClaimBridge.

Do not use `LIM-RPS` as the canonical name for the whole movement runtime. It is
a real historical/research term in the repo, but not the source-of-truth name for
every component.

Diffusion / Flow Correction

Current MotionMixApp has:

DiffusionService
  -> ConditioningEncoder
  -> FlowGenerator1Step

with a temporary 104D CoreML encoder shim. That matches the user's memory of a
conditioning encoder plus one-step flow path. It should not be described as a
verified full multi-step diffusion system.

Convergence

Long-term convergence requires paired data:

• body motion;
• audio;
• N'Ko text/ASR output;
• inscription claims;
• music/output parameters;
• timestamps and confidence.

Only then can a shared body/audio/N'Ko latent be trained and evaluated.

Read Order

1. [comp-core-lumecc-nko-unification.md](comp-core-lumecc-nko-unification.md)
2. [architecture-alignment.md](architecture-alignment.md)
3. [maoe-routing.md](maoe-routing.md)
4. [nko-motion-inscription.md](nko-motion-inscription.md)
5. [parallel-architectures.md](parallel-architectures.md)
6. [convergence-vision.md](convergence-vision.md)