Keyword boosting

The keywords parameter improves transcription accuracy for specified words. Provide the words as a list to improve results for audio that contains them.

Keyword boosting is currently supported only when transcribing Korean speech. With the sommers model, write each keyword according to its Korean pronunciation using Hangul. With the whisper model, keywords may contain Hangul, English abbreviations, and numbers. English letters are case-insensitive.

Name	Type	Description	Example (sommers)	Example (whisper)
keywords	array	List of keywords to boost accuracy	["에스티티", "에이피아이"]	["stt", "에스티티", "위스퍼 V2", "Api"]

caution

• With sommers, write keywords according to their Korean pronunciation. For example, use 에스티티 and 에이피아이 instead of STT and api.
• With sommers, keywords must contain only Hangul. For example, 에스TT, 에스ㅌㅌ, and 에스티티2 are invalid; 에스티티 is valid.
• To use keyword boosting with whisper, set language to ko and transcribe Korean speech.
• With whisper, keywords may contain Hangul, English abbreviations, and numbers. English letters are case-insensitive. Examples include stt, 에스티티, 위스퍼 V2, and Api.
• With whisper, English letters and numbers must form abbreviations whose spelling and pronunciation have clear boundaries, such as STT, CBT, and V2.
• Words such as Agenda, 1on1, and B2B may not benefit from boosting when their written and spoken boundaries are ambiguous.
• Each keyword is limited to 20 characters, with up to 500 keywords.

Usage

Set keywords in the POST request body.

POST request example: Batch STT sample code

Complete example: Batch STT sample code

Example Improvement

When keywords includes ["음성인식"], an unclear pronunciation recognized as "음성의식" may be corrected to "음성인식".

{
  "id": "G6j7wIYxSCCz3gYtQbeYdQ",
  "status": "completed",
  "results": {
    "utterances": [
      {
        "start_at": 1187,
        "duration": 600,
        "spk": 0,
        "spk_type": "NORMAL",
-       "msg": "리턴제로는 음성의식을 하는 회사입니다."
+       "msg": "리턴제로는 음성인식을 하는 회사입니다."
      }
    ]
  }
}