Features

Keyterms Prompting

Boost recognition of names, jargon, and domain vocabulary by passing up to 1000 key terms with the request.

Every domain has words a general model rarely hears: product names, people, acronyms, technical jargon. Keyterms prompting lets you pass those words with the request so the transcriber is primed to recognize them. Supply up to 1000 terms in keyterms_prompt; anything beyond that is dropped.

Quickstart

bash

curl -s -X POST https://api.realtimevoicekit.com/v2/transcripts \
  -H "Authorization: Bearer rtvk_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "audio_url": "https://example.com/standup.mp3",
    "model": "whisper-voicekit-best",
    "keyterms_prompt": ["RealtimeVoiceKIT", "Kubernetes", "Priya Sharma", "OKR"]
  }'

Tips for good keyterms

Use the exact surface form you want in the transcript, including casing: "RealtimeVoiceKIT", not "realtime voice kit".
Prefer multi-word phrases for names and products; they give the model more context than single tokens.
Send only terms likely to occur in this recording. A short, relevant list beats an exhaustive glossary.
Keyterms bias recognition; they do not force a word into the transcript when the audio clearly says something else.

Keyterms prompting works best with the whisper-voicekit-best model.