Korean speech to text API: transcribe Korean audio

To convert Korean speech to text, send the recording to a speech-to-text API with the Korean language code ko and read back the transcript with a start and end time for each word. With Sume STT 1.0, that is one POST /v1/stt-1.0/transcribe request carrying the audio's public HTTPS URL and "language_code": "ko". Leave the code out and the language is auto-detected.

STT 1.0 is specified in the OpenAPI document behind the Sume API reference, which the API reference docs page names as the source for exact request and response shapes, read on 2026-09-27. Speech-to-text API with word timestamps covers polling and webhooks for any language.

How do I transcribe Korean audio?

Pass ko as the language hint; the API reference's own STT request example does the same. The body takes these fields:

• audio_url: the recording at a public HTTPS URL. The reference's examples point at .m4a and .wav files.
• language_code: an optional hint of 2–16 characters, such as ko. Omit it to auto-detect; the result then reports the language_code it found, when available.
• duration_seconds: the recording's length, 1–600. It sizes the usage reservation; omit it and Sume reserves 1 minute.
• segmentation: { "mode": "sentence" } to get sentence segments as well as word timings.

curl -X POST https://api.sume.com/v1/stt-1.0/transcribe \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: interview-ko-007" \
  -d '{
    "audio_url": "https://example.com/audio/interview-ko.m4a",
    "language_code": "ko",
    "duration_seconds": 480,
    "segmentation": { "mode": "sentence" },
    "mode": "async"
  }'

What does the Korean transcript contain?

Once the job is completed, GET /v1/jobs/{id}/result returns these fields. Tokens in words[] can carry a type such as spacing, so count or align only word tokens, and show text rather than joining tokens yourself.

Does it split the Korean transcript into sentences?

Only if you ask. With segmentation: { "mode": "sentence" }, STT 1.0 groups the timed words into sentences on terminal punctuation and splits unpunctuated runs on silence, so a stretch of speech without sentence-ending punctuation is still broken up.

Segments are gapless time ranges over your file: each one's end equals the next one's start, and no audio clips are cut.

How long can a Korean recording be, and what does it cost?

Up to 10 minutes per request: duration_seconds runs 1–600, “Maximum 10 minutes”. Split a longer interview or lecture into parts and shift each part's times by its start offset, as in Transcribe long audio files.

STT 1.0 bills $0.01 per audio minute on API pricing, plus a 5.5% agent fee by default, so a full 10-minute request is $0.10 before the fee.

Can I turn the transcript into Korean subtitles?

Yes. The sentence segments give you timed lines to start from. To burn them onto a video with Sume's captions API, pass each line's text, start, and end as segments and use a Hangul caption style: the Latin styles slam, punch, and tiktok-green reject Korean copy with 400 (caption_hangul_text_latin_style). Korean subtitles API lists the Hangul styles and fonts.

In current code, a caption job refuses a video longer than 60 seconds, so cut a longer video into parts first, as in Add captions to a long video.

For a subtitle file instead, see how to generate an SRT file. To get English from the Korean audio, see translate audio to English.