Overview#
The Silent and Duress Caller module enables emergency services to assist callers who are physically unable to speak freely, individuals experiencing domestic violence, hostage situations, child emergencies, or any scenario where a nearby aggressor makes verbal communication dangerous. When a caller activates silent mode, the AI dispatcher seamlessly transitions into an interaction model designed for constrained communication: either a binary yes/no system controlled by phone keypad tones, or a carefully maintained cover conversation that disguises the call's true purpose while quietly gathering address and situation information for dispatch.
The module is designed with a life-safety bias: when in doubt, the system interprets ambiguous signals as distress and escalates rather than risks inaction.
Key Features#
- Three Interaction Modes: The system supports normal (standard call), yes/no (binary DTMF responses), and pretend (AI adopts a cover conversation persona) modes, transitioning between them based on caller signals
- DTMF Yes/No Interaction: In yes/no mode, callers respond to AI questions by pressing 1 for yes and 2 for no, allowing full triage and address confirmation without the caller speaking a single word aloud
- Cover Conversation Personas: In pretend mode, the AI adopts one of several natural-sounding cover identities, a pizza shop taking an order, a wrong-number recipient, or a telephone survey caller, allowing the caller to remain on the line without arousing suspicion from nearby listeners
- Code-Word Detection: The system recognises a set of community-known duress signals, including "order a pizza" phrases, "red code", "Angel Shot", and wrong-number scenarios with distress markers, as well as personalised safe words configured by the caller's registered profile
- Forgiving Pattern Matching: Detection is intentionally biased toward false positives; misidentifying a genuine pizza order as a duress signal has far lower consequences than failing to detect a genuine emergency
- Address Gathering Under Cover: When operating in pizza-order persona, the AI extracts delivery address, flat/apartment number, and any additional location detail through natural conversation, quietly capturing the location data needed for dispatch
- Transcript Suppression: Cover conversation transcripts are withheld from dispatcher-facing dashboards that could be visible to the caller or a nearby aggressor, preventing accidental exposure of the call's true nature
- P1 Immediate Escalation: As soon as silent mode is confirmed, the incident is raised to P1 priority and an urgent alert is pushed to the dispatcher panel so response units can be deployed without waiting for additional information
- Voice Modulation Guidance: The AI maintains a consistently calm, unhurried vocal tone in pretend mode, designed not to betray urgency to listeners in the caller's environment
Use Cases#
Domestic Violence, Cannot Speak Freely#
A person calls emergency services with an abusive partner nearby. They cannot say anything suspicious. The AI recognises the "order a pizza" phrase and confirms silent mode. The AI becomes a pizza shop, gathers the address through a natural order conversation, and dispatches police silently. The caller remains on the line without the partner suspecting the call was to emergency services.
Child Calling While Hidden#
A child hiding during a home intrusion calls emergency services but whispers "red code" in a barely audible voice. The system detects the duress signal, switches to yes/no mode, and confirms address and intruder presence through single keypad presses. Units are dispatched at P1 without the child speaking further.
Hostage Communication#
A person being held against their will manages to dial emergency services but cannot risk the call being detected. They press 1 and 2 in response to yes/no questions from the AI, confirming their location, the number of people present, and whether they are injured, a full triage without a single spoken word.
Pre-Registered Safe Word#
A user has previously registered a personal safe word in their caller profile during a safety planning process with a support organisation. When they use that word on a call, even in a seemingly normal context, the system recognises it and initiates silent mode with immediate escalation.
Suspected Duress Without Explicit Signal#
A caller is behaving oddly, speaking haltingly, giving inconsistent information, using a strained tone. The audio intelligence layer flags possible stress indicators. The dispatcher can manually trigger silent mode confirmation, prompting the AI to gently offer a yes/no check-in without exposing the question to listeners.
How It Works#
Integration#
The Silent and Duress Caller module operates within the Voice AI session layer and integrates with the DTMF input handler in the telephony transport, the Caller Enrichment module for safe-word lookup, and the Incident Management module for priority escalation. Code-word detection and mode transitions are handled in the edge voice worker, ensuring sub-second response to distress signals. Dispatcher alerts are surfaced through the Operations Dashboard as priority escalation events. All mode transitions, DTMF inputs, and cover conversation transcripts are recorded in the call audit log with appropriate access controls.
Open Standards#
- NENA i3 (NENA-STA-010.3): Silent-mode activation creates a P1-priority incident routed through the NG9-1-1 Emergency Incident Data Object (EIDO) pipeline, ensuring the escalation reaches the Computer-Aided Dispatch system through the standard i3 ESRP pathway without any spoken word from the caller.
- RFC 4733 / RFC 2833 (RTP Payload for DTMF Digits): The yes/no interaction mode relies on DTMF tones carried per RFC 4733 (telephony) and via the W3C
RTCDTMFSenderAPI on WebRTC sessions; digit routing is handled in the silent-caller state machine with digits 1 and 2 mapped to yes and no respectively. - RFC 7852 (Additional Data for Emergency Calls): On silent-mode activation, the NG9-1-1 Additional Data Repository attaches RFC 7852 SubscriberInfo and custom extension blocks, including any pre-registered safe-word profile, to the incident record so dispatchers receive caller context without requiring the caller to speak.
- EN 17128:2020 (Advanced Mobile Location): The PSAP transport layer parses AML SMS bodies conforming to the European EN 17128 standard, giving the dispatcher an automatically derived device location for silent callers who cannot verbalise their address, including in the pizza-order cover conversation where the delivery address supplements or confirms the AML fix.
- NENA-STA-012 / RFC 7865 / RFC 7866 (SIPREC): All mode transitions, DTMF inputs, and cover-conversation audio are captured in tamper-evident SIPREC segments per NENA-STA-012 and sealed with a Merkle anchor, providing the post-incident audit trail required for safeguarding reviews without exposing content during the live call.
- APCO/NENA Call-Taking Performance Standards: P1 silent-mode escalation is subject to the same NENA-mandated 15-second answer and dispatch-alert thresholds as any life-threat priority call; the QA service scores silent-caller handling against these benchmarks.
Compliance#
- Pretend mode transcripts are withheld from real-time dispatcher views to prevent accidental exposure but are retained in the secure audit trail for post-incident review
- Safe-word configuration is stored within the caller's opted-in profile record and is subject to the same data access controls as other sensitive profile fields
- Silent mode activation events and their trigger source are recorded for quality assurance and safeguarding audit purposes
- The module does not store biometric voice data from cover conversations; only the structured call outcome and address data gathered are persisted
Last Reviewed: 2026-04-14 Last Updated: 2026-04-14