Songstark Logo

SPEECHMAP · 3-AXIS AI ENGINE

Five minutes. Three axes read at once.

Linguistic · articulation (DDK) · acoustic — three streams fused into one engine that replaces the traditional checklist screening with a 5-minute speech sample, flagging faster and more accurately which children need to move on to formal assessment. Live inside Dururbarun Co-op’s speech therapy program.

Talk to salesAPI partnership

WHY WE BUILT IT

Delay-visits up 41.5% in 3 years.

10–15% of children aged 2–7 show language-development delay. Korea has only 12,000 licensed SLPs, and the golden window closes fast — while a formal assessment still costs 1.5–2 hours and ₩300–500K.

0-4 delay-visits, 3-year growth (2021→2023)
+41.5%

19,102 → 27,021

early-intervention success (before vs. after age 4)
90% → 60%

golden window < 5

young children with suspected language delay
1 / 10
potential demand vs. licensed SLPs
370K / 12K

HIRA (2021–2023) · ASHA 2020 · domestic market estimates

WHERE WE FIT

The missing seat between checklist and formal assessment.

Checklists are fast but miss 30%. Formal assessments are accurate but run 1.5–2 hours at ₩300–500K. We build the gateway engine between them.

  1. TIER 1

    Checklist screening

    K-DST · SELSI

    10–20 min questionnaire. Subjective, 30% miss rate. Low cost — but the borderline group slips through.

  2. TIER 2 · OUR SEAT

    SpeechMap

    3-axis integrated AI

    5 min of real speech → linguistic · articulation · acoustic auto-analysis. 90% accuracy, ₩20–30K per run, operable by teachers and parents.

  3. TIER 3

    Formal standardized assessment

    PRES · REVT · APAC

    1:1 SLP-administered, 1.5–2 hours, ₩300–500K. Accurate — but time, cost, and access are the walls.

INSIDE THE ENGINE

One utterance. Three axes read in parallel.

Linguistic, articulatory, and acoustic analyses usually run separately. We fuse all three inside one engine and turn 5 minutes of speech into a clinician-reviewable report.

  1. AXIS 1 · Linguistic

    Linguistic

    Post-STT pattern analysis of phonology · vocabulary · grammar · pragmatics. Expressive vocabulary, sentence structure, and response consistency — all in one pass.

  2. AXIS 2 · Articulation

    Articulation (DDK)

    Diadochokinetic-rate analysis measures agility and precision of the articulatory apparatus — catches articulation-disorder signals directly from the waveform.

  3. AXIS 3 · Acoustic

    Acoustic

    Tone · prosody · speech rate · intonation extracted from raw audio. Picks up the grain that STT alone drops — disfluency, distortion, stutter.

end-to-end flow

  1. 015-min speech capture
  2. 023-axis parallel analysis
  3. 03AI fusion report
  4. 04Clinician review · EMR

HOW IT DIFFERS

Four approaches, one engine.

Checklists, formal assessments, ambient-audio AI — each misses something. We fill the seat the others can’t.

  • Checklist

    K-DST · SELSI

    Method
    Questionnaire
    Time
    10–20 min
    Cost / run
    Low
    Accuracy
    30% miss rate
    Operator
    Parent / teacher
    Coverage
    Narrow

  • Formal

    PRES · REVT · APAC

    Method
    In-person assessment
    Time
    1.5–2 hours
    Cost / run
    ₩300–500K
    Accuracy
    High (per domain)
    Operator
    Licensed SLP
    Coverage
    Per-domain, separate tests

  • Ambient AI

    LENA

    Method
    Long-form ambient capture
    Time
    10-hour recording
    Cost / run
    High
    Accuracy
    Strong in natural context
    Operator
    Teacher
    Coverage
    Volume · frequency

  • SpeechMap

    3-axis AI

    Method
    5-min speech · AI
    Time
    5 minutes
    Cost / run
    ₩20–30K
    Accuracy
    90% · 15% miss
    Operator
    Teacher · parent + SLP review
    Coverage
    3 axes integrated

TRACTION

Running in the field.

Dururbarun · KAIST

Clinical partnerships

SpeechMap engine live inside Dururbarun Social Cooperative’s speech-therapy program. Clinical and research design co-reviewed with KAIST faculty.

9 filings

In-house patents

PCC-DP · hallucination-suppressed LLM · 3-minute screening · clinical report — 9 applications filed with KIPO. Paper-oriented R&D track.

HOW IT LANDS

Three channels. One engine.

Same engine — SaaS subscription for B2B, provincial license for B2G, partner API for B2C. Three markets reinforcing each other.

B2B

Therapy centers · clinics

5,000 domestic speech-therapy centers · outpatient clinics

₩50K–500K/month SaaS subscription. Clinician dashboard + auto-report. Target coverage: top 300 centers (10%).

2028 revenue ₩250M

B2G

Education authorities · public

17 metropolitan/provincial + 226 municipal + 103 district offices

Provincial license + teacher-run mode. Scale a province-wide standard screening tool into Seoul-metro and Yeongnam regions.

2028 revenue ₩50M

B2C

Home · partner API

≈2M children aged 2–7

Open the engine API to smart-toy / kids-device / education-service partners. Home AI personalized training (1-min units) beta 2026-04.

2028 subscribers 2,000

WHAT'S NEXT

2026 · 2027 · 2028.

BEP 2026-04 → net-positive 2027-01 → SaMD Class 2 clearance 2028. From gateway to regulated asset.

  1. 2026

    Launch · BEP · AI training beta

    AI personalized training (1-minute units) beta. 20 partner institutions · 1–2 additional patent filings · ₩150M revenue · BEP reached in April.

  2. 2027

    Scale · break-even

    100 institutions · 3 additional patents · B2C expansion. ₩500M revenue · net-positive in January. B2B API partner private beta opens.

  3. 2028

    SaMD Class 2 · global footing

    Software-as-Medical-Device Class 2 clearance · clinical trial + IRB complete · ₩700M revenue · 2,000 subscribers · global launch groundwork.

Pilot · PoC · API.

We take clinical and education PoCs one case at a time. If you want to run real voice data through the 3-axis engine, just say so — no sales PM in between, a core engineer replies within one business day.

Talk to salesAPI partnership