Blog
Guides and insights on AI voice agents, call centers, and calling at scale.
Glossary & Guides
What Is Latency, and Why Does It Matter on Calls?
Published
Most people have experienced a phone or video call where both people start talking at the same time, then both stop, then both start again. Or a call where every reply seems to come just a little too late, making the conversation feel stilted. The cause is usually latency, and it's one of the most important factors in how natural a call feels.
The Definition
Latency is the delay between something being said and the other person hearing it. On calls, it's usually measured in milliseconds (thousandths of a second).
There are two common ways to describe it:
One-way latency, the time it takes for audio to travel from the speaker to the listener.
Round-trip latency, the time it takes for audio to travel to the other person and back again.
How Much Latency Is Acceptable?
The International Telecommunication Union (ITU), the United Nations agency for telecommunications standards, provides guidance on this. Its recommendation, known as G.114, suggests that one-way delay should ideally stay below about 150 milliseconds for most conversations to feel natural. Delays above around 400 milliseconds are generally considered unacceptable for normal conversation.
People rarely notice delays below 150 milliseconds. As delay increases, conversations start to feel awkward, and people begin talking over each other.
Why Latency Happens on Calls
On internet-based calls, latency builds up from several small delays.
Distance
Signals take time to travel, even at close to the speed of light. In fiber-optic cables, signals travel at roughly 200,000 kilometers per second, adding around 5 milliseconds of delay for every 1,000 kilometers. Calls between distant countries naturally have more delay.
Network routing
Data doesn't travel in a straight line. It passes through many routers and networks, each adding a small delay. Inefficient routing, such as data taking a long detour, increases latency.
Packaging the audio
On VoIP calls, audio is converted to digital data, compressed by a codec, and split into small packets, often covering 20 milliseconds of speech each. This process adds a small delay.
Jitter buffers
Because packets can arrive at uneven intervals, phones and systems hold incoming audio briefly in a jitter buffer to smooth it out. This improves audio quality but adds delay.
Network congestion
When a network is busy, packets may wait in queues, adding delay.
Satellite links
Calls routed through geostationary satellites experience much higher latency, since signals travel tens of thousands of kilometers to space and back.
Devices and software
Phones, computers, headsets, and software all add small processing delays. Wireless headsets and busy computers can add noticeable latency.
VPNs and security systems
Routing calls through VPNs or security systems, particularly if they're located far away, can add delay.
How Latency Affects Conversations
Talking over each other
When delay is high, each person thinks the other has finished and starts speaking, only for both voices to overlap.
Awkward pauses
Replies seem slow, making conversations feel stilted or making one person seem hesitant.
Echo becomes more noticeable
Echo, hearing your own voice repeated back, is far more distracting when there's significant delay.
Frustration and misunderstanding
Over a long call, high latency can make conversations tiring and lead to misunderstandings.
Latency and AI Voice Agents
For AI voice agents, there's an additional kind of latency: the time between a caller finishing speaking and the AI starting to reply. This response latency includes network delays plus the time needed to:
- Recognize the caller's speech.
- Decide the caller has finished speaking.
- Generate a response.
- Convert the response into speech.
Since people typically reply to each other within a fraction of a second in normal conversation, AI systems aim to respond quickly enough to feel natural, often targeting replies within about a second. Techniques such as streaming each step and running systems close together help reduce this delay.
Latency vs Jitter vs Packet Loss
Latency is often discussed alongside two related problems:
Jitter is variation in delay, when packets arrive at uneven intervals, causing choppy audio.
Packet loss is when packets never arrive, causing gaps or distorted audio.
All three affect call quality, and they often occur together, especially on congested networks.
How Latency Is Measured
Latency can be measured in several ways:
- Network tests, such as "ping," which measure round-trip time to a server.
- Call quality reports from phone systems, which often show latency, jitter, and packet loss for each call.
- Monitoring tools, which track latency across a network over time.
Reducing Latency
Several practical steps can help:
- Use a wired internet connection rather than Wi-Fi where possible.
- Ensure enough bandwidth for the number of calls being made.
- Prioritize voice traffic on the network, using quality of service (QoS) settings.
- Choose providers with servers close to callers.
- Avoid unnecessary routing, such as sending calls through distant VPNs.
- Use wired or low-latency headsets.
- Keep devices and software updated and avoid overloading computers during calls.
- Monitor call quality to catch problems early.
The Invisible Factor in Every Conversation
Latency is easy to overlook because when it's low, no one notices it at all. But once it rises past a certain point, it changes how conversations feel, making them awkward, tiring, and harder to follow. For businesses, keeping latency low is a key part of delivering clear, natural calls, whether customers are speaking with a person on the other side of the world or an AI voice agent in a data center.