Speech & pronunciation

NSD-Speech-Stack

한국어 음성 인식·합성·발음 평가 통합.

Korean speech recognition, synthesis, and pronunciation assessment.

한국어 STT (자체 학습 모델) + TTS + 발음 평가 자산 통합. 발음 평가 처리시간 60초 → 5–7초 (~10–12× 가속), 평균 confidence 0.929, 평균 발음 점수 81.75. STT 운영 24시간 6,190 요청, 9일 누적 가동.

Korean STT (in-house model) + TTS + pronunciation assessment integrated. Pronunciation latency 60s → 5–7s (~10–12× speedup), mean confidence 0.929, mean pronunciation score 81.75. STT served 6,190 requests over a 24-hour window, 9-day uptime.

Mini-dashboard · indicators only

Verified signals. 추세·비교·상태만 — 절대 수치 일부는 공개 보류.

~10–12×

Pronunciation speedup

60 s → 5–7 s wall-clock

0.929

Mean confidence

10-run benchmark

81.75

Mean pronunciation score

10-run benchmark

6,190

STT requests · 24h

9-day production uptime

Pronunciation latency · before vs after (seconds)

prior60current6

60 s → 5–7 s wall-clock · midpoint 6 s shown

Pronunciation confidence

93%

0.929 mean

10-run benchmark

Higher is better · 0–1 scale shown as %

Components in production

STT 24h · 6,190 production requests · 9-day uptime

Included NSD products

  • NSD-STT-Korean-v1.4
  • NSD-TTS-v1
  • NSD-PopKor-Pronunciation-v1

Pronunciation: 10-run benchmark · STT: 24h production count · 9-day uptime.