Ticket NumberPMD-3930 (FR ???)
Project GoalExpose an open, low-latency WebSocket interface that delivers the live audio of an in-progress call to a partner-specified wss:// endpoint — listen-only for real-time transcription, live captions and in-call compliance, or bidirectional so a partner's automated voice agent can converse with the caller, with barge-in and end-of-speech signalling for natural turn-taking.
Why this project exists ?

Live call audio is currently unreachable to third parties:

  • the Call Control API is a CTI channel that carries call setup and call-state events over WebSocket but does not carry media;
  • Portaone audio is only available post-call, since PortaSIP writes a copy of the RTP stream to disk and a recording server converts it to .wav/.mp3 after the call ends, Portaone with speech-to-text running on that finished recording.
  • the Voice AI Agent (Apifonica) module added in MR130 does real-time AI voice, but the agent is configured on Apifonica's own platform and assigned to a number or extension Portaone — a partner cannot substitute their own STT/model/TTS stack.

Meanwhile WebSocket media streaming is the established interface across the voice API market (Twilio, Vonage, Telnyx, Plivo, SignalWire) so partners wanting real-time voice AI build on a competing CPaaS and leave PortaSwitch doing billing only.

Who are the users / whom we bring value ?
  • External service providers and Add-on Mart partners — the primary consumers; connect OpenAI Realtime, Deepgram, ElevenLabs or their own stack without protocol conversion or their own media servers.
  • CSPs — sell AI, captioning and compliance features without operating media infrastructure.
  • Business customers (PBX Administrators) — point any number or extension at an external AI or compliance service.
  • Contact-centre operators and agents — live agent assist, real-time quality monitoring, automated first-line answering.
  • End subscribers with accessibility needs — live captions and running transcripts, served by listen-only streaming alone.
  • Compliance / data-protection officers — in-call intervention instead of post-call review.
What are the benefits for CSP/ PortaSwitch owner?One platform capability serves three separate wishlist items — live captions (WSHLST-74), sentiment analysis (WSHLST-75) and AI call qualification (WSHLST-203). Keeps AI application development, and its recurring revenue, on the platform rather than on a competing CPaaS. Opens an Add-on Mart category for conversational AI, transcription and real-time analytics with revenue share. Interruption-capable real-time voice is a primary differentiator in competitive UCaaS displacement. Meets accessibility legislation (ADA, EN 301 549, Equality Act) through listen-only streaming alone. Extends compliance from post-call review to in-call intervention.
Target Release
N/A
AreaPortaSIP

Specifications

Definitions, acronyms and abbreviations

AbbreviationDescription
CSPCommunication Service Provider
ISVIndependent Software Vendor — a company selling its own software product, here one built on top of PortaSwitch
ITSPInternet Telephony Service Provider — a provider delivering voice services over IP (SIP trunking, hosted PBX, VoIP lines)
CPaaSCommunications Platform as a Service — e.g. Twilio, Vonage, Telnyx, Plivo, SignalWire
UCaaSUnified Communications as a Service
CTIComputer-Telephony Integration — programmatic call control from an external application
STTSpeech-to-Text (speech recognition)
TTSText-to-Speech (speech synthesis)
RTPReal-time Transport Protocol — carries the media of a SIP call
SIPRECSIP Recording — standard protocol for network-side media forking
CLI / CLDCalling Line Identification (caller number) / Called Number
DIDDirect Inward Dialling number
ACLAccess Control List — controls which users may see or edit a feature
WSSWebSocket Secure — WebSocket over TLS
Barge-inThe caller speaking over playing audio; the agent's playback is stopped immediately so it does not talk over the caller
Media forkingDuplicating a live call's media to a second destination without affecting the call

References

  1. API for computer-telephony integration services — the existing WebSocket call control API, establishing that a control channel exists but carries no media.
  2. Call recording — the current post-call audio path (RTP to disk, conversion after the call, NewCallRecordingIsAvailable event).
  3. What's new in MR130 — Voice AI Agent (Apifonica), the existing single-vendor real-time module.
  4. https://docs.x.ai/developers/model-capabilities/audio/speech-to-speech
New project document ...


Create "Business Requirements"
Create "HLD / LLD"
Create "Test plan"Create "External UAT"Create "Deployment Guide"

Participants ...

Business Requirements

Implementation

Testing