How do you measure the voice quality of a SIP trunk or an 0800 number?
By listening to a real call. Monitor records the audio the test call received and compares it with the expected recording (your IVR greeting or the test tone) using a perceptual model that gives a score from 1 to 5, the MOS-LQO. It measures what a customer would hear, not a calculation made from network figures. That is how call quality monitoring catches a call that connects and sounds bad.
Measured or estimated: what’s the difference?
| Estimated MOS (E-model) | Measured MOS-LQO (Monitor) | |
|---|---|---|
| Where it comes from | Packet loss, jitter, and delay reported by the network | The audio the call received, compared with the reference |
| Standard | ITU-T G.107 (R-factor) | ITU-T P.800.1 (objective listening quality, reference-based) |
| Sees audio cut at the source | No | Yes |
| Sees a wrong message | No | Yes: the comparison with the reference fails |
Both measures have their place. A network estimate is useful for looking at the network; a call that is actually listened to tells you whether the voice got through. See the glossary for MOS, MOS-LQO and POLQA and PESQ, and the guide to MOS and MOS-LQO in practice.
What is a good score?
The MOS scale runs from 1 (bad) to 5 (excellent), as in ITU-T P.800. In practice, around 4 the voice sounds clear, and below 3 people start to complain. Monitor’s default thresholds are:
| Range | Reading |
|---|---|
| 3.8 or higher | Healthy |
| From 3.0 to 3.8 | Working, with the score flagged as below ideal |
| Below 3.0 | Degraded: the test call counts as degraded |
Where does quality show up?
On each circuit’s page, with the score of every test call and the threshold lines; on the dashboard next to the circuit’s status; and in the SLA reports. A drop below the threshold goes through the same retries as any failure before it becomes an alert.
Why is the reference recording needed?
Because a perceptual model compares two signals: what should have been heard and what was heard. During onboarding, 2N5 records the reference for each number. The stored audio is that of the recording or the IVR, never a person’s voice, and it is kept for the period 2N5 sets.
What about jitter, packet loss and delay?
The measured score shows the effect on the voice; jitter and packet loss help find the cause in the network. To see whether the degradation came from the site’s internet link, see internet link monitoring. If the problem is a silent call or one-way audio, the guide to silent calls lists the most common causes.
How does Monitor compare with tools that estimate MOS from the network?
See the comparison with PRTG, which estimates the score from the network, and with VoIPmonitor, which analyzes captured calls, each with the source for its claims.
Frequently asked questions
What is MOS in VoIP and what is a good score?
MOS (Mean Opinion Score) is the 1-to-5 score that summarizes the quality of a voice call, on the ITU-T P.800 scale. Around 4 the voice sounds clear, and below 3 people start to complain. A good score for your circuit is one that stays steady and well away from the alert threshold you agreed.
What is the difference between network MOS and MOS-LQO?
Network MOS is estimated: a model calculates the score from packet loss, jitter and delay, without listening to anything (ITU-T G.107). MOS-LQO is measured: a perceptual model compares the audio that arrived with the reference recording (ITU-T P.800.1). POLQA and PESQ are models of that kind, and scores are only comparable within the same model.
How do I measure the voice quality of a VoIP call?
By listening to the call, not just looking at the network. Place a test call, keep the audio that arrived and compare it with the expected recording using a perceptual model, which gives a score from 1 to 5. It is what Monitor by 2N5 does on every answered test call, around the clock, and it alerts when the score drops below the agreed threshold.
Which measurement model do you use?
A reference-based perceptual listening-quality model that returns MOS-LQO (ITU-T P.800.1). It compares the received audio with the reference audio and returns a score on the MOS scale. Because it uses a reference recorded during onboarding, the result also reveals a wrong message, audio cut off at the source and a silent call.
What are the thresholds?
By default, a circuit is healthy at a MOS-LQO of 3.8 or higher and degraded below 3.0; in between, the call passes with a note. 2N5 tunes the thresholds with you, per circuit, during onboarding, because a customer service line demands more of the voice than an internal test line does.
What happens when quality drops?
A score below the threshold marks the test call as degraded, and it goes through the same retries as any failure, the retest before alerting. If it persists, a quality incident opens and the escalation ladder is notified, with the score of each call as proof for the carrier.