Getting started

Speech Understanding

Extract structured insight from your audio: summaries, sentiment, chapters, entities, speakers, and free-form questions.

Speech Understanding turns a finished transcript into insight you can act on: meeting summaries, sentiment reads, chapter breakdowns, named entities, and answers to any question you want to ask about the recording.

All requests use the base URL https://api.realtimevoicekit.com and authenticate with an API key sent as Authorization: Bearer rtvk_your_api_key. See Single Sign-On for where keys are created and managed.

Two ways to understand audio

The API gives you two complementary tools. Use the analysis flags when you know up front what processing a recording needs, and the prompt endpoint when you want to interrogate a finished transcript on demand.

Approach	How it works	When to reach for it
Analysis flags on `POST /v2/transcripts`	Boolean options on the transcript request. Each requested pass returns its results as a dedicated field on the completed transcript response.	You know at submission time that the recording needs speaker turns, a summary, sentiment, chapters, or entities.
Prompt endpoint `POST /v2/speech-understanding`	Send any question or instruction about a completed transcript and get a grounded answer back.	Custom formats, action items, or any ad hoc question, decided after the transcript exists.

Analysis options at transcript creation

POST/v2/transcriptsCreate a transcript with analysis options

Five boolean flags on the transcript request enable understanding features. All default to false, and each one returns its results on the completed transcript response.

Flag	What it enables	Result field on the transcript
`speaker_labels`	Separates the audio into speaker turns.	`utterances`: array where each entry carries the speaker, its text, and millisecond timing.
`summarization`	Summarizes the recording while the audio is processed.	`summary`: a generated summary string.
`sentiment_analysis`	Analyzes sentiment across the spoken content.	`sentiment_results`: array of per-segment sentiment results.
`auto_chapters`	Segments the recording into chapters.	`chapters`: array of auto-detected chapters.
`entity_detection`	Detects entities such as people, organizations, and locations.	`entities`: array of detected entities.

Each result field is null unless its flag was requested, and a requested pass that found nothing also reads as null, so null-versus-value is all your code needs to branch on.

bash

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/meeting.mp3",
    "speaker_labels": true,
    "summarization": true,
    "sentiment_analysis": true,
    "auto_chapters": true,
    "entity_detection": true
  }'

The transcript processes asynchronously. Poll GET /v2/transcripts/{transcript_id} until status is completed, or pass a publicly reachable webhook_url on the create request and the API will POST a JSON notification to it when the transcript reaches completed or error. The completed transcript then carries the core fields (text, words, utterances, confidence, audio_duration_sec) plus the result field for every flag you requested.

Webhook notification payload

{
  "event": "transcript.completed",
  "transcript_id": "tr_abc123",
  "status": "completed"
}

Ask questions over a transcript

POST/v2/speech-understandingAsk a question about a completed transcript

Send a transcript_id and a prompt, and the API answers using only the transcript content. The transcript must have been created through POST /v2/transcripts on the same account and must be completed.

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": "List every decision made in this meeting and who owns the follow-up for each."
  }'

Response

{
  "id": "su_9f2c31ab",
  "transcript_id": "tr_abc123",
  "model": "default",
  "response": "Three decisions were made. 1) Ship the beta on Friday (owner: Dana). 2) Move the retro to Tuesdays (owner: Sam). 3) Pause the ads experiment (owner: Priya).",
  "created_at": "2026-07-22T14:03:11Z"
}

Answers are grounded: the model is instructed to answer strictly from the transcript and to say so when the transcript does not contain enough information. See the Prompting guide for the full request and response reference.

Speech Understanding

Two ways to understand audio

Analysis options at transcript creation

Ask questions over a transcript

Explore the features

Summarization

Sentiment Analysis

Chapters and Entities

Prompting guide