Skip to main content
November 2026 · Vobiz Team Internet telephony is the transmission of voice calls over an IP network — the internet — rather than the traditional circuit-switched telephone system (PSTN). It converts voice into digital packets, routes them like any other internet traffic, and reconstructs them at the destination. VoIP (Voice over Internet Protocol) and SIP trunking are the most common implementations. Every modern cloud calling platform, CPaaS, and AI voice agent runs on internet telephony at its core.
Key takeaways
  • Internet telephony replaces copper circuit-switching with IP packet routing — the same infrastructure as the internet
  • VoIP is the technology; SIP is the signalling protocol; SIP trunking is the enterprise connectivity product
  • AI voice agents depend on internet telephony — specifically SIP + WebSocket — to intercept call audio and process it in real time
  • Audio quality is determined by codec choice (narrowband G.711 vs wideband G.722/Opus), network jitter, and packet loss
  • Cloud telephony is internet telephony delivered as a managed service with an API — no on-premises hardware required

​
Traditional telephony vs internet telephony

Traditional PSTN (Public Switched Telephone Network): Voice is encoded into a continuous electrical signal (analogue) or a dedicated digital circuit (ISDN/T1/E1). A fixed 64 kbps channel is reserved for the entire duration of the call, even when no one is speaking. Voice quality is consistent but infrastructure is expensive, inflexible, and difficult to scale. Internet telephony: Voice is digitised, compressed into packets, and transmitted over IP. Packets share network bandwidth with other traffic. No dedicated circuit is reserved. The connection is elastic — hundreds of calls can share the same internet uplink. Infrastructure is programmable via API.

​
Core protocols

IP (Internet Protocol): The underlying routing mechanism — packets are addressed and routed independently through the network. RTP (Real-time Transport Protocol): Carries the actual voice audio packets. RTP runs over UDP (not TCP) because voice requires low latency and can tolerate occasional packet loss better than retransmission delay. SIP (Session Initiation Protocol): The signalling protocol that establishes, modifies, and terminates calls. SIP handles the “dial”, “answer”, “hold”, and “hang up” signalling. The voice audio itself travels separately via RTP. WebSocket: Increasingly used in AI voice agent deployments — a persistent full-duplex connection that streams audio bidirectionally between the telephony layer and the AI processing pipeline.

​
How an internet telephony call works

For AI voice agent calls, the architecture adds a WebSocket layer:
The SIP trunk provider is the bridge between the traditional phone network and the IP/WebSocket world.

​
Internet telephony components for businesses

SIP trunk: A virtual phone line delivered over the internet. Replaces physical PRI/T1/E1 lines. Connects a business PBX or cloud application to the PSTN for making and receiving real phone calls. DID (Direct Inward Dialling): A real phone number assigned to a SIP trunk. Callers dial a standard phone number; the call routes over SIP to your application. PSTN gateway: Hardware or cloud service that converts between PSTN signalling (SS7) and SIP for calls that need to reach traditional telephone networks. Media server: Processes RTP audio — codec transcoding, mixing, recording, noise cancellation, VAD. The media server quality determines call audio quality. CPaaS (Communications Platform as a Service): Internet telephony delivered as an API. Vobiz, Twilio, Plivo, and Telnyx are CPaaS providers. You call an API to place a call, receive webhooks for events, and control call behaviour programmatically.

​
Codec choice and audio quality

The codec determines audio bandwidth and packet size. For AI voice agents, codec choice directly affects ASR accuracy. For AI voice agent deployments, Opus or G.722 at the media server significantly reduces ASR word error rates compared to G.711. Vobiz media servers transcode to Opus for AI agent WebSocket streams, maintaining HD quality even when the carrier delivers G.711.

​
Latency and packet loss

Jitter: Variable packet arrival times. If packets arrive 20ms apart on average but sometimes arrive 50ms apart, the reconstructed audio has gaps. Jitter buffers compensate at the cost of added latency. Packet loss: Lost RTP packets result in audio dropouts. At <1%, barely noticeable. At 3–5%, audio quality degrades significantly. At >10%, calls become unintelligible. Packet Loss Concealment (PLC) algorithms in the media server interpolate the missing audio — reducing the perceptual impact. One-way latency target: <150ms is imperceptible; 150–400ms is noticeable but acceptable; >400ms causes echo and conversational awkwardness.

​
Internet telephony and AI voice agents

AI voice agents cannot run on traditional PSTN infrastructure — they require internet telephony:
  1. Programmatic call control: The AI platform must trigger outbound calls via API and receive inbound calls via webhook. PSTN has no API layer.
  2. Real-time audio access: The AI needs the raw audio stream to run ASR. Internet telephony via WebSocket delivers this as a continuous audio stream. PSTN has no equivalent.
  3. Bidirectional streaming: The AI synthesises speech (TTS) and needs to inject it into the call in real time. WebSocket supports bidirectional streaming; PSTN does not.
  4. Low-latency interconnect: AI processing introduces ~800ms–1.5s of latency. Internet telephony networks with regional PoPs minimise the network portion of this latency budget.
Vobiz provides the internet telephony layer — SIP trunking, number provisioning, media server, and WebSocket audio streaming — optimised for AI voice agent deployments.

​
Internet telephony in India

India has a large and growing internet telephony market, governed by DoT and TRAI:
  • Domestic VoIP: Permitted for licensed providers. App-to-app VoIP (WhatsApp, Google Meet) is permitted.
  • PSTN interconnect: VoIP termination to Indian PSTN numbers requires a licensed provider.
  • Commercial outbound calls: TRAI TCCCPR mandates 140-series caller ID for all B2C commercial calls, including those delivered via internet telephony. See India Calling Regulations.
  • Mumbai PoP: Vobiz operates a media PoP in Mumbai — critical for low-latency AI voice agent deployments targeting Indian consumers. Indian-origin SIP traffic processed locally rather than routed through Europe or the US.

​
Internet telephony vs cloud telephony vs VoIP

These terms are closely related but have distinct meanings:
  • Internet telephony: The technology category — voice over IP networks
  • VoIP: A specific implementation of internet telephony using IP protocols
  • SIP trunking: An enterprise connectivity product — a virtual phone line delivered via SIP over the internet
  • Cloud telephony: Internet telephony delivered as a managed cloud service with an API (no on-premises hardware)
  • CPaaS: Cloud telephony with a programmable API layer for custom application development
For most businesses deploying phone call automation or AI voice agents today, “internet telephony” and “cloud telephony via API” are the practical implementation — no hardware, pay-as-you-go, globally addressable.