Con tecnologiaChatGPTClaudeGoogle Gemini
Funziona conGoogle DriveDropboxOneDrive

Guides

Review flagged content

Audit completed transcripts with targeted prompts and build structured moderation reports.

The content_safety report on a moderated transcript tells you what the screening flagged; a review workflow tells you what a human should look at. This guide builds one from two endpoints you already have: POST /v2/speech-understanding for the review itself and DELETE /v2/transcripts/{transcript_id} for cleanup.

Because speech understanding answers are grounded strictly in the transcript, the review cannot invent violations: every finding it reports can be traced back to a quoted line.

Audit a transcript with a prompt

Ask for a review of the completed transcript. Instruct the prompt to quote each concern so a reviewer can verify findings against the transcript text.

bash
curl -X POST https://api.realtimevoicekit.com/v2/speech-understanding \
  -H "Authorization: Bearer rtvk_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "transcript_id": "tr_abc123",
    "prompt": "Review this transcript for harmful, threatening, or sensitive statements. Quote each one, explain the concern, and say none found if the transcript is clean."
  }'

Build a moderation report in JSON

For automated pipelines, set response_format to "json" and define the report schema in the prompt. Every transcript then produces the same keys, ready to route into a review queue.

Structured moderation report
curl -X POST https://api.realtimevoicekit.com/v2/speech-understanding \
  -H "Authorization: Bearer rtvk_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "transcript_id": "tr_abc123",
    "response_format": "json",
    "prompt": "Return a JSON object with keys \"verdict\" (clean or flagged), \"findings\" (array of objects with \"quote\" and \"reason\"), and \"personal_data_mentioned\" (array of strings naming the kinds of personal details spoken aloud, such as phone number or address)."
  }'

The response field contains a JSON string; parse it and branch on verdict. Transcripts that come back clean can flow straight through, while flagged ones go to a human with the quoted findings attached.

Delete after review

When a review concludes a recording should not be retained, delete the transcript. The delete is permanent and returns 204 No Content.

Delete a transcript
curl -X DELETE https://api.realtimevoicekit.com/v2/transcripts/tr_abc123 \
  -H "Authorization: Bearer rtvk_your_api_key"
4.2su 21 recensioni