RealtimeVoiceKIT documentation

The RealtimeVoiceKIT API turns audio and video into accurate, structured text. Transcribe pre-recorded files or URLs, stream live audio over WebSockets, run realtime voice agents, and ask questions over completed transcripts, all through one JSON REST API at https://api.realtimevoicekit.com/v2.

Every endpoint returns word-level timestamps and confidence scores, optional speaker labels, and ready-to-use SRT or VTT subtitles. Transcription supports more than 95 languages with automatic language detection, and every capability is powered by the Whisper VoiceKit model family.

The API authenticates with secret keys that start with rtvk_. Create and revoke keys in the API dashboard; each key's full secret is shown only once, at creation time. Send the key as a Bearer token in the Authorization header of every REST request:

Authorization: Bearer rtvk_your_api_key

A quick way to verify a key is to list the available models:

curl https://api.realtimevoicekit.com/v2/models \
  -H "Authorization: Bearer rtvk_your_api_key"

Create a transcript from any public audio or video URL, then fetch it until status is completed. To transcribe local files, first request a signed upload URL with POST /v2/uploads/sign; the Pre-recorded STT guide walks through both flows.

curl -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/interview.mp3",
    "model": "whisper-voicekit-best",
    "speaker_labels": true
  }'

# Then fetch the result until status is "completed"
curl https://api.realtimevoicekit.com/v2/transcripts/TRANSCRIPT_ID \
  -H "Authorization: Bearer rtvk_your_api_key"

The completed transcript includes the full text, word-level timestamps, per-word confidence, and speaker-labeled utterances when speaker_labels is enabled. Optional flags on the create request unlock speech-intelligence fields in the same response: summarization fills summary, auto_chapters fills chapters, entity_detection fills entities, sentiment_analysis fills sentiment_results, and content_moderation fills content_safety. Each field stays null unless its flag was requested. Subtitles are one request away at GET /v2/transcripts/{id}/srt or /vtt.

Prefer push over polling? Include a public webhook_url when creating the transcript and the API POSTs a JSON notification to it as soon as the job reaches completed or error. Delivery details are in the FAQ.

Five Whisper VoiceKit models cover pre-recorded transcription, realtime streaming, and voice agents. Pick a model per request with the model field. The Models page covers each one in depth.

• Models: choose the right model for accuracy, latency, and cost.
• Account management: accounts, API keys, and team workspaces.
• Billing and pricing: free credit, metered rates, and usage tracking.
• FAQ: quick answers on formats, limits, languages, and support.