What is VoIP? How a phone call travels over the internet
VoIP (Voice over Internet Protocol) means carrying a conversation as data packets instead of over a dedicated telephone circuit. Almost every call you make today is VoIP somewhere along the way; the question is only where.
From circuits to packets
The classic telephone network reserved a fixed channel for the whole duration of a call. Even when nobody spoke, that capacity was held. VoIP does the opposite: your phone or computer samples your voice, compresses it, cuts it into small packets (typically one every 20 milliseconds) and sends them across an IP network, where they share the road with web pages, videos and everything else. At the other end the packets are put back in order and turned into sound.
That change is why internet calls can be so cheap: the network capacity is shared, and an app-to-app call between two people on Wi-Fi costs the provider very little to carry. It is also why VoIP calls are more sensitive to network conditions than an old landline: a packet that arrives late is as useless as one that never arrives.
The two jobs: setting up the call and carrying the voice
Every VoIP system has to solve two separate problems.
Signalling: who is calling whom
Before any voice flows, the two ends must find each other, agree to talk and agree on how. In the standards-based world this is done with SIP, the Session Initiation Protocol defined by the IETF in RFC 3261. A SIP "INVITE" message is the digital equivalent of the phone ringing; it carries a description of the media the caller can handle (codecs, addresses, ports), written in SDP, the Session Description Protocol (RFC 8866). The callee answers with its own capabilities and both sides settle on a common format.
Carrier services such as VoLTE and Wi-Fi calling use SIP inside the operator's IMS core. Business phone systems and many VoIP providers use it too. Consumer apps such as WhatsApp, Signal or FaceTime use their own signalling protocols, but the job is the same.
Media: the voice itself
Once the call is set up, audio travels in RTP packets, the Real-time Transport Protocol of RFC 3550. Each packet carries a sequence number and a timestamp, so the receiver can detect losses and reorder late arrivals. RTP normally runs over UDP rather than TCP: for live speech it is better to skip a lost packet than to wait for it to be resent. When the media is encrypted, the secure variant SRTP (RFC 3711) is used.
Codecs: how voice is squeezed into packets
A codec turns sound into bits and back. The choice of codec decides how natural the call sounds and how much data it consumes.
| Codec | Where you meet it | Audio bitrate |
|---|---|---|
| G.711 | Traditional telephony, many SIP phones | 64 kbit/s, narrowband |
| G.729 | Older business VoIP, low-bandwidth links | 8 kbit/s, narrowband |
| AMR-WB (G.722.2) | "HD Voice" on mobile networks | 6.6 to 23.85 kbit/s, wideband |
| EVS | Newer VoLTE and VoNR networks | 5.9 to 128 kbit/s, up to fullband |
| Opus | WebRTC, many apps | 6 to 510 kbit/s, adaptive |
"Narrowband" codecs reproduce roughly the 300–3,400 Hz range of the old telephone; wideband and fullband codecs carry much more of the voice, which is why app calls often sound clearer than a call to a landline. Opus, standardised in RFC 6716, adapts its bitrate to the network in real time, which helps on unstable mobile connections. How much data these codecs actually use once packet headers are added is worked out in our guide to VoIP data usage.
Getting through home routers: NAT, STUN, TURN and ICE
Most phones and computers sit behind a router that shares one public address between many devices (network address translation, or NAT). That makes it hard for a caller to send packets directly to you. VoIP systems work around it with three standard tools: STUN (RFC 8489) lets a device discover its public address; TURN (RFC 8656) relays media through a server when a direct path is impossible; and ICE (RFC 8445) tries the candidate paths and picks the best one that works. When a call connects but you hear nothing in one direction, a failure in this step is a common cause.
App-to-app calls versus calls to real numbers
It helps to separate three kinds of "internet call", because they behave very differently:
- App-to-app calls (WhatsApp, Signal, FaceTime, Teams…). Both people need the same app. The call never touches the telephone network, is usually free beyond your data, and can be end-to-end encrypted. These apps generally cannot reach emergency numbers.
- VoIP-to-phone calls. A provider connects your internet call to the public telephone network through a gateway, so you can ring any landline or mobile. This is where per-minute charges, caller ID and regulation come back in. The early smartphone calling services of the 2010s worked this way.
- Carrier VoIP. VoLTE and Wi-Fi calling are also VoIP, but run by your mobile operator with your normal number, inside its own managed network.
What decides whether a VoIP call sounds good
Bandwidth is rarely the problem: a voice call needs far less than a video stream. What matters is latency (how long a packet takes to arrive), jitter (how much that delay varies) and packet loss. The ITU's recommendation G.114 notes that most users are satisfied when one-way delay stays below about 150 ms. Our guide to call quality problems covers how to diagnose and reduce each of these.
In short
VoIP is not a product but a method: voice cut into packets, set up by a signalling protocol and carried in real time. Whether a particular VoIP call is free, reaches any number, shows your caller ID, works for emergencies or is private depends on which of the three families above it belongs to. The rest of this site takes those questions one at a time.