Features

Sentiment Analysis

Read the emotional tone of a recording and find where it shifts.

Sentiment analysis tells you how a conversation felt: whether a support call recovered after a rough start, which parts of a sales demo landed, and where a meeting turned tense. Enable the sentiment_analysis flag at creation to get per-segment results back on the completed transcript, and use prompts when you want a higher-level read.

Enable sentiment analysis at creation

Set sentiment_analysis: true on the transcript request. Pairing it with speaker_labels is useful for conversations, because the completed transcript then carries per-speaker utterances you can reason about.

Create a transcript with sentiment analysis

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/support-call.mp3",
    "speaker_labels": true,
    "sentiment_analysis": true
  }'

The completed transcript then includes a sentiment_results array with one entry per analyzed segment, carrying the segment's text along with its sentiment label, timing, and confidence values. The field is null unless the flag was requested (and when the pass found nothing to analyze).

Go further with a prompt

Per-segment labels are the raw material; conclusions usually need context. Ask POST /v2/speech-understanding for the sentiment view you need, from a one-word overall label to a timeline of tone shifts.

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": "Describe the overall sentiment of this call, then note any moment where the tone shifts and why."
  }'

Question	Prompt pattern
Overall tone	In one word, was the overall sentiment of this call positive, neutral, or negative? Then justify in one sentence.
Per speaker	For each speaker, describe their overall sentiment and quote one line that shows it.
Escalation check	Did the customer sound frustrated at any point? If so, quote the moment and describe what triggered it.
Resolution check	Did the tone improve by the end of the call? Answer yes or no, then explain briefly.

Sentiment as JSON

For dashboards and pipelines, set response_format to "json" and define the object shape in the prompt so every call returns the same keys.

Structured sentiment

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 JSON with keys \"overall\" (positive, neutral, or negative), \"confidence_note\" (string), and \"turning_points\" (array of short strings)."
  }'