Jitter, latency and packet loss are the three network measurements that decide whether a VoIP call sounds clear or terrible. A VoIP call is a stream of small packets, about 50 a second. Voice is unforgiving: a web page can wait a second for a missing piece, but a conversation cannot.
This guide explains what each measurement means, what it does to a call, the targets commonly used, and how to find and fix problems. If you only want an estimate for your own numbers, use the VoIP call quality (MOS) calculator.
Latency in VoIP calls
VoIP latency is the time a packet takes to get from one end to the other. For voice, what matters is one-way delay, which is roughly half the round-trip time you see with a ping.
When delay grows, the call starts to feel like a walkie-talkie: people talk over each other, then both stop and wait. ITU-T recommendation G.114 advises keeping one-way delay under 150 ms for most applications.
The delay in a call is more than the network. It is a sum of several parts:
| Part of the delay | Typical size (example) |
|---|---|
| Network travel time | 50 ms |
| Codec processing and packetisation (G.711, 20 ms packets) | 20 ms |
| Receiving jitter buffer (about twice the jitter) | 20 ms |
| Total one-way delay | 90 ms |
A good connection leaves plenty of room under 150 ms. The room disappears quickly with long routes, VPNs that send traffic the long way round, or satellite links.
VoIP jitter
VoIP jitter is how much the delay varies from packet to packet. Packets are sent evenly, but they do not arrive evenly: some are held up behind other traffic.
Voice has to be played out smoothly, so the receiving phone uses a jitter buffer: it holds packets for a moment and plays them out at a steady pace. A bigger buffer absorbs more jitter but adds delay. A buffer that is too small lets late packets arrive too late to use, and they are discarded. So jitter turns into either delay or loss.
Common causes are congestion, Wi-Fi, and routes that change mid-call. Phones with adaptive jitter buffers grow and shrink the buffer to match, which usually keeps calls clear until the jitter gets large.
VoIP packet loss
VoIP packet loss is the share of packets that never arrive in time to be played. The listener hears missing syllables, clipping and robotic sound. Codecs differ in how well they hide it. Some hide small amounts of random loss reasonably well, but loss that comes in bursts is much harder to conceal.
Packets can be lost on the network, dropped at a congested router, or discarded by the phone because they arrived too late (which is how jitter becomes loss). Wi-Fi interference is a frequent source.
What they do to call quality together
The ITU's E-model combines delay, loss and the codec into a quality score. This is the simplified form used in our MOS calculator, for two narrowband codecs: G.711 with packet loss concealment, and G.729A. The jitter column is the measured jitter; the model assumes a jitter buffer of about twice that value.
| Scenario | Latency | Jitter | Loss | G.711 MOS | G.729A MOS |
|---|---|---|---|---|---|
| Good connection | 50 ms | 10 ms | 0% | 4.36 | 4.01 |
| High latency | 150 ms | 10 ms | 0% | 4.27 | 3.79 |
| Very high latency | 300 ms | 30 ms | 0% | 3.19 | 2.51 |
| High jitter | 50 ms | 40 ms | 0% | 4.33 | 3.96 |
| 1% loss | 50 ms | 10 ms | 1% | 4.27 | 3.84 |
| 2% loss | 50 ms | 10 ms | 2% | 4.17 | 3.67 |
| 5% loss | 50 ms | 10 ms | 5% | 3.83 | 3.21 |
| 10% loss | 50 ms | 10 ms | 10% | 3.30 | 2.61 |
A few lessons stand out. Loss does more damage than most people expect, and it hurts the more compressed codec most. Latency is forgiving until it crosses roughly 150 ms and then falls off quickly. And a score above 4 is good, 3.6 or below is where some callers start to complain, and 3.1 or below is poor. These are estimates for planning, not measurements.
Commonly used targets
For a good voice call, targets often quoted in vendor QoS guidance are:
- one-way latency of 150 ms or less;
- jitter of 30 ms or less;
- packet loss of 1% or less.
They are guidelines, not cliffs. Quality falls gradually, and different codecs tolerate different amounts.
How to measure jitter, latency and packet loss
- Ping and traceroute show round-trip time and where the delay builds up. Tools such as mtr combine both and show loss per hop.
- iperf3 in UDP mode sends a steady stream and reports the jitter and loss it sees, which is close to what voice experiences.
- Softphone and handset statistics often show jitter and loss for a live call.
- Provider dashboards may report call quality using RTCP statistics.
- A test call at the busy hour tells you more than one at 7 am.
How to find the cause
- Is it one user or everyone? One user points to their device or connection. Everyone points to the office network or the provider.
- Wired or Wi-Fi? Plug in a cable and test again. If the problem disappears, the cause is the wireless link.
- Does it happen when someone uploads? Calls that break up during big uploads or backups point to congestion or bufferbloat (see below).
- Run mtr to the provider during a bad moment and look for the hop where delay or loss begins.
- Check the device. An overloaded computer running a softphone can add jitter and drop audio.
- Try a different codec or packet interval, if your system allows it.
- Collect evidence before you contact the provider: times, call IDs, and mtr results.
Usual causes and fixes
| Symptom | Common cause | What helps |
|---|---|---|
| Echo, talk-over | High latency | A closer provider region, fewer hops, avoid VPNs that route calls the long way round |
| Choppy audio | Jitter | Wired connection, QoS, a less congested link |
| Dropouts, robotic voice | Packet loss | Fix the congested link or poor Wi-Fi, give voice priority |
| Good alone, bad when others work | Congestion | QoS, more bandwidth, limit big uploads at peak times |
| Fine until someone uploads | Bufferbloat | Smart queue management on the router, or shape traffic just under the line speed |
Bufferbloat in one paragraph
Many routers hold large queues of packets. When a link is saturated, the queue fills and every packet, including your voice packets, waits behind the rest. Latency can jump by hundreds of milliseconds under load even though the connection is fast. The fix is a router that manages its queue sensibly (look for smart queue management, often using fq_codel or CAKE), or shaping traffic a little below the real line speed so the queue stays short.
QoS and DSCP
QoS means marking voice packets (commonly with the DSCP value EF) and having the router give them priority over bulk traffic. It only helps if every router in the path honours the marking, so it works best inside your own network. On the open internet, the marking is usually ignored, which is why a clean, uncongested connection matters more than any marking.
Bandwidth is a separate question
A call can use a thin link if nothing else is using it. Congestion is what hurts. Work out how much you need with how much bandwidth VoIP needs and the codec bandwidth calculator, then check the connection with the steps above.
FAQ
What is a good latency for VoIP?
ITU-T G.114 recommends keeping one-way delay under 150 ms for most uses. Many calls stay comfortable well below that.
Can Wi-Fi cause jitter and packet loss?
Yes. Wi-Fi is a common cause, because interference and weak signals add delay variation and dropped packets. Wired connections are steadier for voice.
How is MOS related to jitter, latency and packet loss?
MOS is a score from 1 to 5 for call quality. It can be estimated from latency, jitter, packet loss and the codec. Our MOS calculator does this.
What is the difference between latency and jitter?
Latency is how long a packet takes to arrive. Jitter is how much that time varies from packet to packet. A link can have high latency but low jitter, and still carry a clear call.
Backend engineer who has built VoIP platforms on Asterisk, with years of Laravel and AWS work behind them. More about us