The MOS score (Mean Opinion Score) rates voice quality from 1 (bad) to 5 (excellent), and in VoIP it is usually estimated from network measurements: latency, jitter and packet loss, plus the codec in use. As a rule of thumb, aim for one-way latency under 150 ms, jitter under 30 ms and packet loss under 1% to keep MOS around 4.
The MOS scale
MOS comes from listening tests defined by the ITU (Recommendation P.800), where people rate calls on a five point scale: 5 excellent, 4 good, 3 fair, 2 poor, 1 bad. Networks cannot ask listeners about every call, so monitoring tools estimate MOS from measurements, typically with the ITU E-model (G.107), which produces an R-factor from 0 to 100 that maps onto the MOS scale.
| Estimated MOS | R-factor | User satisfaction |
|---|---|---|
| 4.3 and above | 90 to 100 | Very satisfied |
| 4.0 to 4.3 | 80 to 90 | Satisfied |
| 3.6 to 4.0 | 70 to 80 | Some users dissatisfied |
| 3.1 to 3.6 | 60 to 70 | Many users dissatisfied |
| Below 3.1 | Below 60 | Nearly all users dissatisfied |
The codec sets the ceiling. G.711 can reach a MOS of about 4.1 to 4.4, while compressed codecs such as G.729 top out lower, around 3.8 to 4.0, even on a perfect network. The trade offs are covered in G.711 vs G.729, and each transcoding step in the chain lowers the score further.
Latency
Latency is the time a voice packet takes from mouth to ear. ITU Recommendation G.114 suggests keeping one-way delay below 150 ms for most applications, and treats more than 400 ms as unacceptable for general network planning. Above roughly 250 ms people start talking over each other. Long international chains, satellite links and jitter buffers all add delay.
Jitter
Jitter is the variation in packet arrival time. The receiving side uses a jitter buffer to smooth it out, which adds delay; packets that arrive after the buffer has played out are discarded and count as lost. Keep average jitter under about 30 ms. Spikes during busy hours usually point to congestion on a link in the chain.
Packet loss
Lost packets cause clipped words and robotic sound. Under 1% loss is usually fine with G.711 and packet loss concealment; above 2 to 3% users notice, and compressed codecs suffer sooner because each packet carries more speech. Bursts of loss hurt more than the same amount spread evenly.
Measuring quality on wholesale routes
In wholesale voice you rarely see the whole path, so combine several sources:
- RTCP and RTCP-XR reports from your session border controller or switch give jitter, loss and round trip time per call on your leg.
- Test call platforms place calls to real handsets and score audio end to end, including the destination network.
- Indirect signs in CDRs: a drop in ACD or a rise in short calls on one route often means people hang up because audio is bad. See monitoring route quality with CDRs.
Remember that your measurement covers only the media path you can see. If audio passes through several carriers, a good MOS on your leg does not guarantee quality at the handset, so confirm with end to end tests as described in how to test a VoIP route.
Common fixes
- Prefer routes with fewer hops and no unnecessary transcoding.
- Agree a codec list with the carrier and avoid forcing G.729 where bandwidth is not a problem.
- Mark voice traffic for priority on your own network and check the internet link to the carrier for loss at peak hours.
- For audio that works in one direction only, follow one-way audio and dropped calls troubleshooting.
Audio quality is one of the main differences between premium and standard routes. If you need better quality on a destination, VoIPMatcher can source VoIP termination with the codec and route type you require: request a VoIP route.
Frequently asked questions
What is a good MOS score for VoIP?
A MOS of 4.0 or higher is generally considered good, and 4.3 or above is very good, close to what G.711 reaches on a clean network. Between 3.6 and 4.0 some users notice problems, and below about 3.1 nearly everyone is dissatisfied. Compressed codecs lower the highest score a call can reach.
How is MOS calculated in VoIP?
Real MOS comes from people rating calls in listening tests. Monitoring tools estimate it instead, usually with the ITU E-model, which combines delay, jitter, packet loss and the codec into an R-factor from 0 to 100 and converts that to the 1 to 5 MOS scale. Different tools can give slightly different scores.
What is the difference between latency and jitter?
Latency is how long a voice packet takes to travel from speaker to listener. Jitter is how much that time varies from one packet to the next. High latency makes people talk over each other; high jitter forces a larger jitter buffer, which adds delay, or causes late packets to be dropped, which sounds like packet loss.
What is a jitter buffer?
A jitter buffer holds incoming voice packets for a short time and plays them out at an even pace, smoothing variations in arrival time. A larger buffer absorbs more jitter but adds delay, and adaptive buffers resize themselves as network conditions change. Packets that arrive after their play time are discarded and count as lost.
What causes poor VoIP call quality?
The usual causes are packet loss, jitter and high latency on a congested link, long carrier chains with many hops, transcoding between codecs, and compressed codecs used where bandwidth is not a problem. On wholesale routes, falling ACD or rising short calls on one route often reveal audio problems before anyone complains.
Need a route?
Send the destination, route type and target rate. We reply with what we can deliver.