Overview#
A duty of care operator calls a traveler to conduct a welfare check during a developing security situation. The call is recorded, transcribed in real time, and analysed for sentiment. The operator hears the traveler saying she is "fine", but the sentiment analysis flags an elevated stress score and an unusual speech pattern. The supervisor sees the summary flag and initiates a follow-up action. A potential issue surfaces that a verbal "fine" would have concealed.
The Duty of Care Telephony module provides that layer of analytical depth on every welfare call, turning voice interactions into documented, analysable records.
Key Features#
- VoIP configuration and call management for duty of care operations.
- Outbound call initiation and call record tracking.
- Call recording with secure storage and playback.
- AI-powered call transcription with full transcript generation.
- Sentiment analysis to assess caller emotional state.
- Key point extraction and AI-generated call summaries.
- Named entity recognition identifying people, places, and organisations mentioned in calls.
- Language detection and transcription confidence scoring.
- Programmable API access for all telephony operations.
Transcription Features#
| Feature | Description |
|---|---|
| Transcript | Full call transcription |
| Summary | AI-generated summary |
| Sentiment | Sentiment analysis |
| Key Points | Extracted key points |
| Entities | Named entities (people, places, organisations) |
| Language | Detected language |
| Confidence | Transcription confidence score |
Use Cases#
Duty of care operators conduct recorded welfare checks with automatic transcription, creating a documented record of every call without requiring manual note-taking during the conversation.
Security supervisors review AI-generated sentiment analysis from welfare calls to identify risk indicators that verbal responses might not capture, enabling proactive intervention when a traveler shows signs of stress or duress.
Post-incident documentation teams generate AI summaries of duty of care calls for supervisor review and case documentation, compressing hours of call recordings into structured records that can be reviewed in minutes.
Compliance and legal teams maintain auditable records of all telephony interactions within duty of care workflows, with complete transcripts, sentiment data, and metadata providing a defensible record of every operator-traveler call.
Integration#
- Connects with duty of care incident management for call-related incident tracking.
- Integrates with voice assistant capabilities for enhanced call features.
- Feeds into duty of care audit logging for complete call audit trails.
Open Standards#
- SIP (RFC 3261): The module uses SIP for all call signalling, including REGISTER for trunk authentication, INVITE for call setup, BYE for teardown, re-INVITE for hold/resume, and REFER (with Replaces) for blind and attended call transfers.
- SDP (RFC 4566): Session Description Protocol is used within SIP re-INVITE messages to negotiate hold and resume state via
sendonlyandsendrecvdirectives. - ITU-T G.711 / G.729: The configured SIP codecs (μ-law, A-law, G.729) comply with these ITU-T standards for PCM and compressed voice encoding over RTP.
- RFC 2833 (DTMF over RTP): DTMF tone signalling is configured as
rfc2833in SIP sub-account provisioning, ensuring interoperable keypad events across carrier trunks. - ITU-T E.164: Outbound call destinations and DID numbers are expected and documented in E.164 format throughout the call initiation and CDR recording APIs.
- CALEA (Communications Assistance for Law Enforcement Act): A dedicated service layer implements the CALEA lawful-intercept request lifecycle, including seven-year encrypted recording retention and full audit trails for access to call recordings.
- CJIS Security Policy (FBI): A dedicated CJIS service enforces clearance-level access controls on recordings, transcripts, and incident data, with audit logging that meets FBI Criminal Justice Information Services requirements.
- WebVTT (W3C): The transcription pipeline captures and stores WebVTT-format captions returned by the Whisper speech-to-text model, enabling time-aligned transcript playback.
Last Reviewed: 2026-02-05 Last Updated: 2026-04-14