Speechlytics documentation
Speechlytics turns recorded and live conversations into searchable transcripts, structured analytics, quality scores and real-time guidance.
This documentation is for developers integrating with the Speechlytics REST and WebSocket APIs: submitting audio, streaming live calls, receiving and interpreting results, connecting call sources, and decrypting payloads.
Choose a workflow
| Workflow | Use it when | Start here |
|---|---|---|
| Pre-recorded speech | You have an audio file, a video, or a downloadable recording URL. | Submit your first transcription |
| Live speech | You need transcript segments and intelligence while the call is happening. | Start a live session |
| Push delivery | You want results delivered to you rather than polling. | Webhooks and call events |
| Provider integration | Calls should arrive automatically from Microsoft Teams, Aircall, FrontSpin, 3CX or SFTP, or results should be written to HubSpot. | Integrations |
| Reading results | You want transcripts, scores, topics, sentiment, QA results and agent analytics. | Retrieving results |
| Payload security | You want responses, events and webhooks encrypted on top of TLS. | How encryption works |
Platform capabilities
Ingestion
- Batch transcription from base64 audio, a direct media URL or a YouTube link
- Scheduled SFTP collection, and provider integrations for Microsoft Teams meeting recordings, Aircall, FrontSpin and 3CX
- Audio extraction from video sources
- Live transcription over WebSocket with a shareable viewer PIN
Transcription
- Whisper-based recognition with a per-account model choice, including an Irish fine-tuned model
- 100 languages plus automatic detection
- Mono and stereo processing, or transcription of only one side of a stereo call
- Per-account audio pre-processing: noise reduction, volume, silence handling, minimum duration and welcome-message removal
- Optional AI polishing of the transcript, automatic for live calls, with revert
Intelligence
- Summaries and structured analysis using a Sales or Customer Support profile, optionally written in the call's own language
- Topics, sentiment, emotion, intent, risk and business signals
- Keyword and phrase scoring and scope filtering
- Custom QA questions with pass/fail verdicts and transcript evidence
- Agent-level analytics and trends
- Transcript search by meaning and across languages (semantic search)
- Sensitive-content masking for recorded and live conversations
- Live keyword detection, alert rules, conversation metrics, sentiment and intent, conversation state, summaries, agent coaching and translation
Delivery and security
- REST polling, a call-events WebSocket and an account webhook
- HubSpot call activities with recordings and summaries on the CRM timeline
- Optional AES-256-GCM encryption of every response, event and webhook
- Two-factor sign-in, account lockout and rate limiting
API conventions
The API uses JSON and is rooted at your deployment's base URL. Examples use https://api.example.com; replace it with the URL supplied for your environment.
Authorization: Bearer <token>
Content-Type: application/json
- Enums are integers. The API reference shows enum values as
2 - Processed; send and read the number (2). - Responses may be encrypted. Every authenticated response carries
Encryption-Data: true|false. See How encryption works. - Optional sections are optional. Many analytics fields appear only when the feature is enabled for the account and the audio supports it. Tolerate missing fields.
Read Authentication and API basics for tokens, two-factor sign-in, rate limits and error handling.
Next steps
- Authenticate and submit audio.
- Read the feature guide to understand account-controlled processing.
- Use the complete API reference for request and response schemas.
Most analytics, live-intelligence and integration capabilities are enabled per account. A valid request can still return 403, or omit optional result sections, when the corresponding feature is not enabled. Contact your Speechlytics representative to change account capabilities.