What Acceptable Latency for VoIP?

Every millisecond counts in a voice conversation. A delay of just a few hundred milliseconds can turn a smooth conversation into one filled with interruptions, repeated questions, and people talking over each other, when the latency is over 200 milliseconds. For businesses that rely on VoIP for sales, customer support, or AI voice agents, those small delays can quickly translate into frustrated customers and missed opportunities.

So, what counts as acceptable latency for VoIP? In this guide, we break down the thresholds that you should know about, how to choose the right VoIP provider and identify network issues and what to do to ensure good performance for your customers.

What is VoIP?

Voice over Internet Protocol or VoIP are phone calls that are made over the internet, rather than using traditional phone systems. VoIP converts your analog voice into compressed digital data packets. Once converted, these packets are sent across the internet to the recipient, where it is then reassembled back into a clear, audible sound.

Some common examples of the VoIP technology are the most common apps available to be downloaded on smartphones, computers and laptops such as WhatsApp, FaceTime, Microsoft Teams and Zoom, that is all done online.

VoIP is commonly used for international calls by individuals. Businesses however could use VoIP to replace their existing communication tools through a cloud-based technology for internal and external use such as conferencing, smart call routing for customer support. It basically provides the caller with the ability to use a business phone number from anywhere in the world.

What is latency? What causes a round trip delay?

Latency is the time it takes your voice to travel from your mouth to the other person's ear. On an old landline that delay was so small you never thought about it. With VoIP calls that is handled over the internet, this is different because your voice doesn't travel as a steady signal down a wire. It gets chopped into small digital packets, sent across the internet one after another, and reassembled at the other end.

It is important to note that every step in that journey adds a little time and the total sum is your latency. latency affects all business industries, not just telecommunications.

What is good latency?
Anything under 50 milliseconds is considered to be good latency or in gaming terms, a ping. Most of the delay comes from three things.

1. Distance: a call to someone across the country has to cover real miles, so an overseas call will always lag a little more than a local one.
2. Equipment in between: every router and switch the packets pass through adds a slight lag.
3. Software on each end: that adds the rest, as it packages your voice up and unpacks it again.

How to tell if latency is your problem?

High latency has a recognisable set of symptoms, and spotting them tells you whether the delay is what's actually hurting your calls or whether you're chasing the wrong thing. Watch for these signs:

1. You hear an echo, your own voice coming back to you a beat after you speak.
2. There's a talk-over problem, where both people keep starting at the same time because neither hears the other quickly enough to take turns.
3. Conversations have awkward pauses, that half-second of dead air after someone finishes before the reply lands.
4. On video calls, the lips don't match the words and there is a mismatch between audio and video. And the audio is choppy or garbled, which usually means latency is showing up alongside jitter or packet loss.

What is acceptable latency for VoIP calls?

Acceptable latency for VoIP is a delay of 150 milliseconds or less from the moment you speak to the moment the other person hears you. At this threshold, a call sounds normal and the delay is almost unnoticed. However, when the delay exceeds this minimal threshold, then conversations are often disrupted and start to feel broken through pauses with people talking over each other.

How is 150ms an acceptable latency?
The International Telecommunication Union (ITU) writes rules that the world's phone networks follow and their guideline for voice set up standards called the G.114, where they suggest that anything between 0-150 milliseconds is deemed as acceptable for most user applications.

However, this often refers to a one-way number, which is the time taken for your voice to make a single trip to the recipient or another speaker. But when you run a test on your own line, the tool almost always gives you a round-trip number, the time for a signal to go out and come back, which is double the time taken from a one-way delay.

Therefore, the 150ms threshold is roughly 300 milliseconds for a round-trip conversation, making you sometimes believe that your conversation is failing, when it isn't in reality.

This table gives you both, side by side, so whichever number your tool reports, you can read it straight off:

‍

The 150 ms figure is the general benchmark, but the real network latency tightens depending on the kind of call. A normal person-to-person business call is barely noticeable up to that 150 ms one-way line. Video calls are can be harder to ignore as delays can become very obvious when there is a mismatch between the audio and the video of a speaker.

A busy contact center delivering high customer experience always wants to stay under 100 ms, since even a small delay repeated across hundreds of back-to-back calls adds up to slower handling and more people talking over each other.

When a person is talking to an automated system, the whole exchange has to feel instant or it reads as robotic, and the delay budget has to cover not just the network but the time to understand the caller and generate a reply.

That's why an AI agent needs far more leeway than a human call does, and why the TTS layer has to respond quickly without too much processing time. This is the reason the number that's comfortable for your sales team may be too loose for a voice bot answering your main line in a contact center.

How to test your VoIP latency?

A VoIP latency test is any check that measures how long a signal takes to reach a destination and return, and the simplest one is already built into your computer. It's called ping. To test VoIP latency, you can use available online tools that are designed to conduct dedicated VoIP latency and speed tests that simulate real call conditions. This should ideally be tested during peak usage hours, using a wired Ethernet connection in the real world.

A simple way is to open a command prompt and ping the address you're calling through, whether that's your VoIP provider's server or the far site. Send at least 100 pings rather than the default few, so one lucky or unlucky result doesn't skew the picture. The average time is your round-trip latency, so halve it for the one-way number to compare against the 150 ms line.

The gap between the fastest and slowest ping tells you how much jitter you have. Jitter refers to the variation in the delay when receiving data packets. When you speak into a VoIP phone, your voice is broken into small data packets that travel across the internet.

When this is completed, take a look at these three things:

1. If you'd rather not read raw ping output, you may use available browser-based VoIP tests like the OnSIP latency and jitter test that accurately shows latency, jitter, and loss together in one click.
2. For a business system where the trouble comes and goes, a monitoring tool like PingPlotter watches the connection over time and points to the exact spot on the route where the delay is creeping in.
3. If you're running voice agents rather than person-to-person calls, the same principle applies, and there's a separate walkthrough on how to test AI voice agents end to end.

Pro tip: Attempt to test from both ends of the connection if possible. A single test would only show you one path at one moment, and the route back isn't always the same as the route out.

How to reduce VoIP latency to acceptable limits?

Once you've tested your latency on VoIP calls and you've come to the conclusion that your numbers are high, the fix usually comes from one of these ways to minimise latency and network jitter.

1. Fix your hardware and internet service provider:
   switch to Ethernet, upgrade an old router, and turn off SIP ALG (Session Initiation Protocol Application Layer Gateway). It is good practice to also check your network latency with your internet service provider. In businesses, this could be what directly affects in providing a good overall customer experience.
2. Prioritise voice traffic:
   Most business phone connections carry voice, email, file downloads, and backups all at once, and when the line gets busy your call packets wait in line behind everything else. A setting called QoS (Quality of Service) tells your network to always let voice packets go first. On a shared connection this is the single biggest win, because it fixes the delay that congestion causes without touching anything physical.
3. Manage bandwidth usage and settings:
   Leave about 20% of your bandwidth free, match your codec to your connection, and turn on an adaptive jitter buffer. This should reduce latency and processing delay for VoIP calls.
4. Choose a better codec:
   A codec is the software that squeezes your voice into data and unpacks it at the other end, and different ones handle bad connections very differently. If your connection sometimes drops voice packets, the Opus codec is the most useful as it can rebuild a lost packet from the one next to it. If your connection is solid and you just want the best sound, G.711 is the simple high-quality choice. Most modern phones and apps support Opus, so it's usually the safe default for VoIP calls.
5. Improve the path:
   Latency is tied to distance and routing, so a provider with servers closer to you, and a wired connection instead of Wi-Fi, both shave real time off. To achieve persistent latency, Fiber is the most stable option.

What is jitter buffer & packet loss?

Jitter is the uneven spacing between packets when some arrive late and others on time.

In VoIP systems, your phone tries to smooth this packet arrival time by holding incoming packets in a small buffer and releasing them evenly. However, this buffer can only absorb so much. When the unevenness gets too big, these packets show up too late to use and get thrown away, and you hear the result as choppy, robotic, or broken-up audio due to a variable delay.

Packet loss is when some of those packets never arrive at all. A little bit is invisible, because the software on the receiving end guesses what was in the tiny gap and fills it. But there are acceptable limits to this as well. When there are too many packet losses, you get the dropped-out words and half-sentences.

Here's the part most guides skip, and it's the most useful thing to understand: a call can survive high latency far better than it survives high jitter or packet loss.

So if your calls sound bad but your average persistent latency looks fine, then you should stop considering average latency as the issue, it's because of jitter and packet loss. Keeping the jitter index under 30ms and the packet loss under 1% should be the ideal goal to have (Nextiva report).

What is a good MOS score?

MOS, or Mean Opinion Score, is a single number from 1 to 5 that rates how good a call actually sounds, where 1 is unusable and 5 is flawless. It started as a score real people gave after listening to test calls, and today, the software estimates it automatically by combining your latency, jitter, and packet loss into one figure. It's the number you'll see on most VoIP monitoring dashboards and provider spec sheets, which makes it handy when you're comparing options.

• A MOS of 4.0 or higher is considered to be of good quality and this is the level that you want for contact center business calls.
• Between 3.5 and 4.0 is usable but noticeably imperfect.
• Below 3.5, people start straining to follow the conversation.
• Achieving a 5.0 as the Mean opinion score is only theoretical, and even a perfect network connection can only give a score of around 4.4 as the codec itself sets a ceiling.

Therefore, a real-world target is 4.0 to 4.4, not 5. If a provider quotes you a MOS, 4.0+ plus is a good score to measure.

Where low latency matters for AI voice agents

Everything discussed above is about the network carrying a call between two people.

If you're building an AI voice agent, latency gets tighter, because on top of the network delay the system also has to hear the caller, think of a reply, and speak it back, and all of that has to happen inside the same budget before the conversation feels slow. Audio quality has to be perfect, network traffic has to be managed and voice quality needs to be seamless for customer satisfaction.

Murf Falcon is built as a sub-800 ms text-to-speech engine so the voice-generation part of the pipeline leaves enough room for the network to stay inside the thresholds on this page, and the latency benchmarks show where that time goes.

If you're weighing platforms to build voice agents, the network numbers here and the processing numbers there are two halves of the same budget, and both have to add up.

‍