VoIP (Voice over Internet Protocol) replaces analog telephony with internet-based voice. Learn how it works, the main types, and how to pick a provider.
VoIP stands for Voice over Internet Protocol. It is a technology that transmits voice calls over the internet instead of traditional copper phone lines. A VoIP system converts your voice into digital data packets, sends those packets across an IP network, and reassembles them as audio on the other end.
You probably use VoIP every day. WhatsApp calls, Zoom meetings, and FaceTime audio all run on it. Businesses use it to replace desk phones, run call centers, and build voice features into software. The concept is simple. The quality of the result depends on the network underneath it.
VoIP is a category of voice transmission, not a single product. Any call that travels as data packets over an IP network counts as VoIP telephony. The term covers consumer apps, business phone systems, and the programmable voice infrastructure that developers build on. People also call it IP telephony, internet calling, or voice over internet.
VoIP technology works with a VoIP number, which looks and dials like any other phone number but routes over the internet instead of the public switched telephone network (PSTN).
There are three ways to make a VoIP call:
All three methods do the same job. They capture your voice, digitize it, and hand it to an IP network for delivery.
Every VoIP call follows the same path. Your microphone captures analog sound. A codec encodes that sound into digital data. The system splits the data into small IP packets, each stamped with addressing information. Those packets travel across the internet, often taking different routes. At the destination, the receiving system reassembles the packets in order, decodes them, and plays the audio.
This happens fast. A well-built network completes the round trip in a fraction of a second. A poorly built one adds delay you can hear, which is why infrastructure matters more than the concept itself.
VoIP call flow
Two protocols do most of the work. SIP (Session Initiation Protocol) handles signaling. It sets up the call, negotiates the session, and tears it down when you hang up. RTP (Real-time Transport Protocol) carries the actual voice packets during the call. SIP is the phone ringing. RTP is the conversation.
Codecs sit between your voice and the wire. Each one trades bandwidth for audio quality differently.
| Codec | Bitrate | Character |
|---|---|---|
| G.711 | 64 kbps | Uncompressed, standard call quality |
| G.722 | 48-64 kbps | Wideband, HD voice |
| Opus | 6-510 kbps | Adaptive, high quality at low bandwidth |
For developers, VoIP is programmable. This Python example uses the Telnyx Call Control API to place an outbound call, speak a message with text-to-speech, and hang up.
import os
import telnyx
from dotenv import load_dotenv
load_dotenv()
# Initialize the client with the new SDK pattern
client = telnyx.Telnyx(api_key=os.getenv("TELNYX_API_KEY"))
# Place an outbound VoIP call
response = client.calls.dial(
from_=os.getenv("TELNYX_PHONE_NUMBER"),
to="+15551234567",
connection_id=os.getenv("TELNYX_CONNECTION_ID"),
)
call_control_id = response.data.call_control_id
print(f"Call initiated: {call_control_id}")
# When the call is answered, speak a message
client.calls.speak(
call_control_id=call_control_id,
payload="Hello. This call was placed programmatically over VoIP.",
voice="female",
language_code="en-US",
)
# End the call
client.calls.hangup(call_control_id=call_control_id)Full example with webhook handling, error handling, and setup instructions: make-outbound-phone-call-python.
Build voice into your productThe Telnyx Voice API runs on a private, owned network with co-located AI inference, delivering sub-500ms latency where multi-vendor stacks average around 1,000ms.
Explore the Voice APIThe PSTN routes calls through dedicated circuits on physical copper and fiber lines. VoIP routes calls as packets on shared IP networks. That single difference drives everything else, from cost to scalability.
| Factor | VoIP | PSTN |
|---|---|---|
| Infrastructure | Internet connection, software | Dedicated copper lines, switches |
| Cost | Lower, no line rental per user | Higher, per-line hardware and fees |
| Scalability | Add users in software | Install new physical lines |
| Features | Routing, recording, AI, APIs | Voice and basic forwarding |
| Call quality | Depends on network conditions | Consistent but narrowband |
| Remote work | Works anywhere with internet | Tied to physical location |
The PSTN still has a place in some deployments, and carriers worldwide are phasing it out at different speeds. For a full comparison, read PSTN vs VoIP.
VoIP services come in several forms. The right one depends on whether you want a finished product, a bridge for existing hardware, or raw programmable infrastructure.
A fixed VoIP number is tied to a physical address, the way a landline is. Businesses use fixed numbers for local presence and regulatory compliance. A non-fixed VoIP number has no address requirement and can be issued anywhere, which suits remote teams and global operations. Either way, the process to get a VoIP number takes minutes rather than the days a landline install requires.
A VoIP network is the collection of servers, gateways, and transport paths that carry voice traffic as IP packets. It replaces the circuit-switched PSTN with packet switching. On the PSTN, a call reserves a dedicated circuit for its full duration. On a VoIP network, packets share bandwidth with other traffic and get routed dynamically. Packet switching is more efficient and far cheaper, but it makes call quality a function of network engineering.
Several components make a VoIP network run:
A VoIP carrier operates this infrastructure and interconnects with the PSTN and other carriers. Here the market splits. Some carriers own their network end to end. Many resell capacity from other operators, which adds hops, middlemen, and latency to every call. When something breaks on a resold network, the provider files a ticket with someone else. When it breaks on an owned network, the provider fixes it.
Voice traffic is real-time, so a VoIP network has strict performance requirements:
| Metric | Target | Effect when missed |
|---|---|---|
| Latency | Under 150ms one way | Talkover, awkward pauses |
| Jitter | Under 30ms | Choppy, distorted audio |
| Packet loss | Under 1% | Dropped words, robotic sound |
| Bandwidth | ~100 kbps per call | Degraded or failed calls |
Quality of Service (QoS) settings help meet these targets. QoS tells routers to prioritize voice packets over bulk traffic like file downloads, since a late voice packet is a useless voice packet. On a private backbone, the operator controls prioritization across the entire path. On the public internet, voice packets compete with everything else. This is why network ownership shows up directly in what a call sounds like.
The commercial case for VoIP starts with cost. Businesses switching from legacy telephony save on line rentals, hardware, long-distance charges, and maintenance contracts. [TODO: verify the 40-60% cost savings figure and the 90% international traffic figure against a citable source before publish. Do not publish either stat unsourced.]
Sub-500ms
Latency with co-located VoIP and AI infrastructure vs ~1,000ms on multi-vendor stacks
Beyond cost, the advantages compound:
Most VoIP providers explain the technology the same way. The differences show up in the infrastructure behind the sales page. Use these criteria to evaluate any VoIP carrier:
Data sovereignty deserves a mention too. When one provider controls the full path from telephony to AI inference, voice data stays on that network instead of transiting third parties. For regulated industries, that is often the deciding factor.
Ready to build on VoIP that runs on an owned network?Get sub-500ms voice latency, global numbers, and one bill instead of six.
Start with the Telnyx Voice APIRelated articles