Back to Language as Infrastructure
working paper2026Script-native ASR manuscript

Living Speech: Script-Native Automatic Speech Recognition for N'Ko

This is the direct ASR pillar: speech should be recognized into the native script instead of being routed through a Latin-script bridge. The paper focuses on the anchor recognition result and why script-native recognition changes what the model is actually learning.

Paper workspace

Live draft structure

working-draft

Artifacts

Draft PDF

Older anchor draft. Useful for reading the ASR argument, but not the latest live calibration truth.

Open artifact

Final split-paper render

Final split-paper artifact for the script-native ASR anchor.

Open artifact

Editable source

Draft PDF exists. It should be refined against the current live-ASR boundary before release.

Source anchors

nko-brain-scanner/paper/current/paper2_living_speech.tex

nko-brain-scanner/paper/final/03-script-native-asr-anchor/paper.tex

Method tags

script-native ASRspeech-to-N'Kodeployment gap

Ingest intersections

asrnkomandingspeechon-device

Status

Drafted; anchor result consolidated into the flagship.

Key claims

01

Direct speech-to-N'Ko is the core target, not transliteration after the fact.

02

A script-native decoder changes the error surface.

03

On-device deployment exposes failure modes that offline CER alone hides.

Public reading note

Drafted; full release should follow stable evaluation packaging.

Standard skeleton

What this paper must keep proving

Schema

problem

Speech systems often route low-resource languages through Latin text before returning to the native script.

method

Recognize directly into N'Ko and measure the error surface without a transliteration bridge.

implementation

CTC decoder over N'Ko symbols, on-device export path, replayable packet capture.

data

Training/evaluation corpora plus live-packet review lanes. Current live truth still requires human labels.

evaluation

Offline CER, live packet gating, and direct transcript review.

references

Whisper, wav2vec-style transfer, CTC, low-resource ASR, Manding language resources.

openQuestions

Whether the 20 percent CER offline model can become the direct live recognizer after frontend parity and label review.

Checkpoints and references

Proof chain

paperpending

Claim checkpoint

central-claim slot

Every central claim must point to a proof anchor or remain labeled as speculative.

implementationpending

Implementation checkpoint

implementation-map slot

Every method should identify the code path, harness, schema, or protocol that embodies it.

experimentpending

Evidence checkpoint

evidence-manifest slot

Every reported result should point to run IDs, packet IDs, data snapshots, commits, or review artifacts.

external-referencepending

Reference checkpoint

references slot

Every external claim should resolve to a cited paper, benchmark, standard, or documented prior system.

paperpending

Release checkpoint

release-gate slot

Every PDF needs a named condition before it can move from draft to citation-ready.