How does VoIP work?

VoIP works by carrying a phone call as digital data over an IP network instead of over the dedicated copper circuit a traditional line holds open for the length of a call.[1] Your voice is sampled, compressed, and cut into a stream of small packets, each addressed like any other piece of internet traffic, and the receiving device puts those packets back in order and plays them as sound.[1][3][12] Because the audio is just data, the same broadband connection that carries your email and video also carries your calls, and a call can start on a laptop or a desk phone and still reach an ordinary phone number.[1][11]

Two jobs on every call

Every VoIP call runs two separate processes at the same time. One is signaling, the work of finding the other party and setting up, changing, or tearing down the call, which is handled most often by the Session Initiation Protocol, or SIP.[2] The other is media, the actual sound, which travels in its own stream using the Real-time Transport Protocol, or RTP, with a sequence number and timestamp on each packet so the receiver can rebuild the audio in the right order.[3] Keeping these two apart is what lets a call negotiate which codec both sides speak before a single word of audio moves.

From your voice to the far end

Between those two jobs sits a fixed sequence that repeats on every call: your device registers with the phone system, a codec digitizes and compresses your speech, RTP ships the audio across the network, and the far end buffers, reassembles, and decodes it.[2][3][9] A call meant for a mobile or a landline takes one more step, passing through a media gateway that converts the IP audio into the signal the public phone network expects.[1][11] The sections below walk through that sequence one stage at a time.

A VoIP call begins with registration, the step where your phone or app tells the phone system what network address to reach it at.[2] SIP handles this by letting a device publish its current location to a registrar, so the system knows where to send an incoming call even when the user moves between a desk, a home office, and a phone on cellular data.[2] When you dial, SIP sends an invitation to the other party, negotiates which audio formats both ends support, and then rings, connects, or ends the call.[2]

This stage decides whether a call can be reached at all and which codec the audio will use before any sound is sent, so it is worth getting right. If you run your own phone system, keeping SIP registration healthy, with credentials and network paths that stay open, is what prevents the common failure where a phone shows as online but never rings. Confirming that both ends share at least one codec during setup also heads off calls that connect with no audio.

The next stage converts the sound of your voice into a digital stream a network can carry. A codec samples the audio at a fixed rate, turns each sample into a number, and compresses the result so it uses less bandwidth.[6] The long-standing baseline is ITU-T G.711, which samples speech 8,000 times a second and produces a steady 64 kilobits per second, the same quality a traditional phone line delivers.[4][6] G.729 compresses the same speech to roughly 8 kilobits per second for links where bandwidth is tight, at a modest cost in fidelity.[6]

The codec you use is a direct trade between sound quality and bandwidth, so the choice matters most on constrained or crowded connections. The Opus codec, designed for interactive audio, adapts on the fly from 6 kilobits per second up to 510 kilobits per second and includes packet-loss concealment, which is why many softphones and browser-based tools use it.[5] For calls that must cross onto the public phone network, an uncompressed path such as G.711 is often the safer choice, because heavy compression can distort the audio during conversion.[6]

Once your voice is digitized, the device slices the stream into small packets and sends them across the network, typically wrapping about 20 milliseconds of audio into each one.[3][6] Those packets travel using RTP, which adds a sequence number and a timestamp to every packet so the receiver can play the audio back in the right order and at the right pace.[3] RTP rides on UDP rather than TCP, which means a lost packet is not retransmitted, because a late repeat of a 20 millisecond slice of speech would arrive too late to be useful.[3]

Sending audio in small, independent packets is what makes a brief network problem cost only a sliver of speech rather than the whole call. It also sets the bandwidth a call needs, because the headers that IP, UDP, and RTP add to every packet push a single G.711 call noticeably above its raw 64 kilobits per second, so sizing a connection means counting concurrent calls rather than looking at one call in isolation.[6]

As the packets leave your device, they are routed across the IP network toward the other party, and on most business and home connections they first have to get past a NAT router and a firewall. Network address translation hides many devices behind one public address, so the private address your phone advertises inside a SIP message is not reachable from outside, which can leave a call connected with one-way audio or no audio.[10] The Interactive Connectivity Establishment framework, known as ICE, works around this by using STUN to discover a device's public address and TURN to relay the media when a direct path is blocked, then picking the best route the two ends can actually use.[10]

Because voice is sensitive to delay and loss, this stage is also where traffic prioritization earns its keep. Marking voice packets so routers handle them ahead of bulk data, and reserving enough upstream bandwidth for the number of simultaneous calls, is what keeps audio clean on a link that is otherwise busy with downloads and video.[9]

At the receiving end, packets rarely arrive at perfectly even intervals, so the device holds them briefly in a jitter buffer before playing them.[9] Jitter is the variation in packet arrival timing, and without a buffer that unevenness turns into choppy, broken audio.[9] The buffer collects a small backlog, reorders anything that arrived out of sequence, and hands a steady stream to the codec, which decodes the packets back into the sound the listener hears.[3][9]

Sizing the buffer is a balance, because a larger buffer hides more jitter but adds delay, and voice is sensitive to delay. A common target keeps jitter under about 30 milliseconds, beyond which callers start to notice the audio degrading.[9] If a packet never arrives, the codec conceals the gap rather than waiting for it, since a late replacement would land after its moment in the conversation has passed.[3] Watching jitter and loss on the receiving side is usually the fastest way to explain why a call sounded rough.

A call between two VoIP endpoints can stay on IP networks from start to finish, but a call to a mobile or a landline needs one more stage. The audio passes through a media gateway that converts the IP packets into the standard signal the public switched telephone network uses, and converts the return audio the same way.[1][11] This is why a VoIP service can keep your existing number and still reach anyone, whatever technology sits on the other end.

The gateway is also where regulation attaches. A service that lets you both place and receive calls to the public phone network over broadband meets the FCC definition of interconnected VoIP, which carries duties an app-only calling service does not.[11] For anyone choosing a provider, whether calls reach the regular network is the line that separates a full phone replacement from a service that only connects to other users of the same app.

Three network conditions decide whether a VoIP call sounds clear: latency, jitter, and packet loss.[8][9] Each traces back to the fact that voice is real-time data sharing a network with everything else, and each has a rough threshold worth watching.

Latency

Latency is the one-way delay between speaking and being heard, and it climbs with distance, slow links, and heavy processing along the path. ITU-T Recommendation G.114 sets the widely used target for one-way transmission time, and independent guidance puts the comfortable ceiling at about 150 milliseconds, above which callers begin talking over each other.[7][8] Delay between roughly 150 and 400 milliseconds is usable but noticeable, and past 400 milliseconds a conversation becomes hard to hold.[8]

Jitter

Jitter is the variation in how evenly packets arrive, and the jitter buffer on the receiving side absorbs a small amount of it.[9] Keeping jitter under about 30 milliseconds generally keeps audio smooth, and the practical fixes are prioritizing voice traffic and relieving congestion on the connection.[9]

Packet loss

Packet loss means some audio never arrives, and because voice runs over UDP those packets are not resent.[3] Modern codecs conceal small amounts of loss, but sustained loss produces clipped or robotic speech, so a link that drops packets under load needs its capacity or its traffic priorities examined before call quality will hold.[9]

VoIP is a method for carrying calls, not a single app or a specific device, and a few distinctions keep the mechanics clear.

It is not one protocol doing everything

Signaling and media are separate jobs. SIP sets up the call and RTP carries the sound, so a problem where calls connect but have no audio usually sits with the media path or a firewall, not with the part that made the phone ring.[2][3]

It is not the same as your cellular voice

Mobile calls on a modern network often travel as an IP voice technology the carrier runs inside its own system. It shares the packet idea with VoIP but is provisioned by the carrier, not set up as a business VoIP account, and it is not the interconnected VoIP service the FCC rules describe.[11]

It is not automatically a regulated phone service

Whether a VoIP service carries phone-company duties depends on whether it reaches the public phone network. An app that only calls other users of the same app is not held to the same 911 and consumer rules as a service that lets you place and receive calls to any number.[11]