Networks and origination
| Term | What it means |
|---|---|
| PSTN | Public Switched Telephone Network — the global circuit-switched / IP voice network that terminates real phone numbers. |
| VoIP | Voice over IP — voice that travels as packets over data networks instead of circuits. |
| SIP | Session Initiation Protocol (RFC 3261) — the signaling layer that starts, modifies, and stops voice sessions. |
| RTP | Real-time Transport Protocol — the media layer that carries the audio packets after SIP sets up the call. |
| SRTP | Secure RTP — RTP encrypted with DTLS-SRTP keys that the endpoints negotiate during signaling. WebRTC requires it. |
| DTLS-SRTP | Datagram TLS that negotiates SRTP keys. It is the standard WebRTC media-encryption handshake. |
| ICE / STUN / TURN | The NAT-traversal toolkit for peer-to-peer media. ICE picks the candidate path. STUN discovers public addresses. TURN relays media when peer-to-peer is blocked. WebRTC uses all three. |
| SDP | Session Description Protocol — the offer/answer body inside SIP/WebRTC that says “I speak Opus and PCMU on UDP/16384”. |
| DID | Direct Inward Dial — a phone number that routes inbound calls into a specific extension or trunk. To “buy a DID” means to rent a phone number. |
| DDI | The European synonym for DID. |
| Toll-free number (TFN) | Numbers such as 800/888. The callee pays the per-minute cost. |
| E.164 | The international number format: +<country><number>, max 15 digits. Always quote and store numbers in E.164. |
| CNAM | Caller-ID Name — the string that the receiving phone shows. It is distinct from the calling number. |
| CLI / Caller-ID | The number that the caller presents to the called party. Regulatory rules apply (US STIR/SHAKEN, EU caller-ID spoofing laws). |
| ANI | Automatic Number Identification — the billing number on the receiving side. 911 and toll-free services use it. |
Trunks and carriers
| Term | What it means |
|---|---|
| SIP trunk | A logical connection between a voice platform and a carrier (or PBX). It has channels, codecs, auth credentials, and an IP allowlist. |
| CPS | Calls Per Second — the pacing limit that a trunk accepts before it throttles. Carriers price by CPS. |
| Channels | The number of concurrent in-flight calls that a trunk can hold. CPS × ACD ≈ steady-state channel usage. |
| SBC | Session Border Controller — a hardened SIP/RTP proxy at the edge of a network. It handles NAT, security, and transcoding. |
| PBX | Private Branch eXchange — an enterprise call switch. It sits between phones and trunks. |
| Origination | Outbound calls, from your platform to the PSTN. |
| Termination | The carrier’s term for delivery of your outbound call to its destination. The same word also means the call’s hangup. |
| Number portability (LNR / LNP) | The ability to keep a phone number when you change carriers. It adds a routing lookup before the call leaves your network. |
| DTMF | The 0–9, *, # tones. There are three signaling modes: in-band (audible), RFC 2833 (RTP events), or SIP INFO. Use RFC 2833 where possible. |
Codecs and media
| Term | What it means |
|---|---|
| G.711 µ-law (PCMU) | An 8 kHz, 64 kbit/s codec. It is the North-American PSTN standard. It applies lossy log compression to 14-bit linear samples. |
| G.711 A-law (PCMA) | The same as µ-law, but with a different log curve. It is the EU/ROW PSTN standard. |
| Opus | A modern wideband codec, 6 kbit/s – 510 kbit/s. It is the WebRTC default. It sounds much better than G.711. |
| G.722 | A 16 kHz wideband codec at 64 kbit/s. Many mobile networks use it as “HD voice”. |
| AMR / AMR-WB | Mobile-network codecs. They are rare on PSTN handoff. |
| Linear16 / PCM16 | Uncompressed 16-bit PCM. ML/AI services usually want this format. |
| Transcoding | Conversion from one codec to another mid-call. It costs CPU and adds ≥ 20 ms of latency. |
| Jitter | Variation in packet inter-arrival time, measured in ms. Jitter > 30 ms typically requires de-jitter buffering. |
| Jitter buffer | An adaptive queue that smooths jitter at the cost of latency. A bigger buffer gives better audio and more delay. |
| Packet loss | The % of RTP packets that never arrive. Loss > 1 % is audible. Loss > 3 % is unintelligible. |
| PLC | Packet Loss Concealment — it synthesizes plausible audio to mask a missing packet. Most modern codecs include it. |
| MOS | Mean Opinion Score, 1.0–5.0. It estimates subjective quality. A score ≥ 4.0 is “toll quality”. |
| R-factor | An objective ITU E-Model score, 0–100. It drives an estimated MOS. |
| VAD | Voice Activity Detection. It suppresses silence frames or triggers barge-in. |
| AEC / AGC / NS | Acoustic Echo Cancellation / Auto Gain Control / Noise Suppression — the three pillars of the WebRTC audio-processing pipeline. |
| Comfort noise | Synthetic background hiss inserted during silence. It keeps the line from sounding “dead”. |
Transports — how the audio actually travels
A modern voice platform must know these four transports:| Transport | What it is | Where it’s used |
|---|---|---|
| SIP + RTP | Classic VoIP. Signaling travels on SIP (TCP/UDP/TLS). Media travels on RTP/UDP. It is the carrier-facing protocol. | Carrier ↔ gateway trunks. |
| WebRTC | Browser-native voice/video. Signaling sends SDP over WebSocket. Media uses DTLS-SRTP over UDP, with ICE for NAT traversal. Media is always encrypted and is peer-to-peer by default. | Browser softphones, click-to-dial, agent dashboards, embedded contact-centre widgets, integrations with LiveKit / Daily / Chime / Twilio Media Streams / Vapi. |
| WebTransport | An HTTP/3 alternative for browser ↔ server bidirectional streams. It uses the same underlying QUIC, with no peer-to-peer and no ICE. | Browser apps that want a low-overhead control plane multiplexed with media on the same QUIC connection. |
| QUIC | A UDP-based encrypted multiplexed transport. Native SDKs use it directly to talk to the gateway. | Server-side Python / Go / Rust / Java / .NET workloads. |
- Use SIP/RTP for the trunk side (carrier ↔ gateway). The PSTN does not speak WebRTC.
- Use WebRTC for the last mile to humans in browsers or vendor SDKs (LiveKit, Daily, Chime, Vapi, Twilio Media Streams).
- Use QUIC / WebTransport for server-side or browser control workloads. Choose them when the audio source is already-decoded PCM (for example, a TTS pipeline) and you want minimum framing overhead.
Call lifecycle and patterns
| Term | What it means |
|---|---|
| A-leg / B-leg | A-leg = caller side; B-leg = callee side. A bridge connects them. |
| Ringback tone | The “ring ring” that the caller hears before answer. The local switch or the far end (early media) generates it. |
| Early media | Audio that the far end sends before the SIP 200 OK. Carriers use it for “the number you have dialed…” intercepts. |
| Bridge | Connect two legs so audio flows both ways. |
| Park | Hold a call without a bridge — keep the line open with music or silence. |
| Music on hold (MOH) | Audio that plays to a parked call. You can set it via URL or stream. |
| Transfer (blind / attended) | The industry terms for the handoff of a call to another destination. Not implemented — out of scope on this platform. |
| Conference | The industry term for a mix of three or more legs into one audio stream. Not implemented — out of scope on this platform. |
| Whisper / Coach | The industry terms for one-way supervisor audio. Not implemented — the platform’s Barge is unrelated; see the next row. |
| Barge (AI barge-in) | The headline AI-network unblocker. While an AI agent speaks, the gateway listens for the human caller’s voice on the inbound leg. When the voice crosses the configured energy / VAD / semantic threshold, the gateway gates the AI’s outbound audio. The human is then heard immediately. Without barge-in, AI voicebots feel deaf: the human must wait for the AI to finish before the AI acknowledges them. Set it per call via auto_barge_in and barge_in_patience_ms, and per trunk via auto_bargein_mode (energy / vad / semantic / off). |
| Barge-in patience | The grace window in milliseconds between detection of the human’s voice and the gate on the AI. A window that is too short stops the AI on a cough. A window that is too long makes the AI talk over the caller. Typical values: 150–350 ms. |
| IVR | Interactive Voice Response — menus driven by DTMF or speech. |
| ACD | Automatic Call Distribution — routing rules that send inbound calls to queues, agents, and skill groups. |
| AMD | Answering-Machine Detection — a heuristic classifier that says “human” vs “voicemail” within the first ~2 s of audio. |
Routing and dialplan
| Term | What it means |
|---|---|
| Dialplan | The rules engine that decides what happens to a call: hang up, park, play, bridge, or hand to AI. |
| LCR | Least-Cost Routing — pick the cheapest carrier per destination prefix. |
| Failover routing | Try carrier A. On 503 Service Unavailable, try carrier B. Trip a circuit breaker after N consecutive failures. |
| Geo-routing | Choose a regional gateway based on caller location, latency, or compliance. |
| Whitelist / Blocklist | Static allow/deny lists per trunk. Carriers often require them for outbound caller-ID. |
| Inbound rule | What to do when a PSTN call arrives at one of your DIDs: REJECT, PLAY-AND-HANGUP, NOTIFY-AND-HANGUP, or HANDLE-AI. |
Compliance vocabulary
| Term | What it means |
|---|---|
| STIR/SHAKEN | The US/CA framework for caller-ID attestation. Every outbound INVITE carries an Identity JWT signed by your carrier or platform. The attestation level (A/B/C) controls how trustworthy the receiving carrier marks your call. |
| TCPA | The US Telephone Consumer Protection Act. It regulates outbound dialing — quiet hours, abandoned-call rate caps, and consent requirements. |
| Do-Not-Call (DNC) | Federal + state lists. DNC suppression must run before every dial. |
| GDPR / CCPA | Privacy regimes that govern call recordings and CDR storage. They affect retention windows. |
| HIPAA | US health-info privacy. Healthcare deployments require a BAA + encrypted recording storage. |
| PCI-DSS | Card-data privacy. If your IVR collects card numbers, the IVR step must redact DTMF from recordings (“DTMF masking”). |
| E911 / 112 | Emergency calls with the location attached as PIDF-LO. It is mandatory for any platform that lets users dial out. |
| CALEA / Lawful Intercept | Government wiretap obligations for carriers. The upstream carrier honours them, not the gateway. |
Where this maps to ClutchCall
This table maps each section above to the page that implements it on this platform:| Concept area | Lives in |
|---|---|
| Trunk lifecycle, codec preferences, inbound rules | Admin → Trunks |
| Outbound origination, bulk dialer, hangup, barge | Public RPC — see Method IDs |
| Dialplan actions (park, MOH, playback, transfer) | ExecuteDialplan and bucket actions |
| WebRTC vendor adapters (Browser, LiveKit, Daily, Chime, Twilio, Vapi) | Platform → WebRTC Vendors |
| Compliance headers (STIR/SHAKEN, PANI, PIDF-LO) | Platform → Regulatory |
| MOS, jitter, packet loss, CDRs | Platform → Telemetry |
The four numbers that bound a voice deployment
Every sizing or capacity discussion comes back to these four numbers:- CPS sustained (per trunk and aggregate).
- Concurrent channel ceiling.
- End-to-end latency (caller mouth → callee ear, in ms).
- MOS under typical packet-loss conditions.