What Is MOS (Mean Opinion Score)?

MOS began as a genuinely human measurement. Panels of listeners rated recorded audio samples from one to five, the ratings were averaged, and the mean opinion score was the result. That method is impractical for a contact center placing thousands of calls a day, so modern platforms estimate the same scale from the network conditions that are known to predict it. The number is a prediction of human judgement rather than a record of one, and it is accurate enough to be extremely useful.

Its value in an outbound or inbound operation is that it settles arguments. Agents say the line is bad. Network teams say the network is fine. Customers say they could not hear. Without a measurement, everyone is describing an impression. With a per-call score, the conversation becomes which calls, on which route, at which times, for which agents.

What Drags the Score Down

• Packet loss. Missing audio, heard as dropouts and clipped words. The single most damaging factor.
• Jitter. Uneven arrival timing, heard as choppy or robotic speech.
• Latency. Delay, heard as both parties talking over each other and then apologising.
• Codec. A narrowband codec caps achievable quality regardless of how clean the path is.
• The last few metres. Wireless networks, shared home connections and cheap headsets, which sit outside the platform entirely.

Why Averages Hide the Problem

Call quality problems are rarely uniform. They cluster on one route, one carrier, one agent's connection or one interval of the day. An average across all calls buries a cluster of unusable conversations inside a comfortable overall figure, and the business only discovers the issue through complaints or a drop in outcomes. Scoring each call individually, and alerting on the low ones while the shift is still running, is what turns audio quality from a retrospective explanation into something a supervisor can act on.

How DialerBee Handles Call Quality

In DialerBee's operations center, mean opinion score is captured per call, and low-MOS alerts surface on the wallboard while the shift is running. The same view carries an ops summary across the platform services, each with its own health state, so you find the failing part before a customer tells you which one it was, along with every call in progress across the platform listed while it is still up. A call trace follows one call end to end by its identifier, which is the answer to what happened on that call rather than a theory about it. Alerts are configurable for a NOC, with snooze so a known issue stops shouting and a history of what fired and when, and remediation actions are recorded in an audit.

On the floor itself, the supervisor tools wallboard carries a MOS card for live audio quality, so when agents say the line is bad the wallboard says whether they are right. Because the numbers that matter are per call rather than per day, a supervisor can move from a low score to the call behind it. Where your calls are carried is your own decision: BYOC connects your own SIP trunks from any SIP-compatible provider, with multi-carrier routing by destination, cost or quality, failover between trunks, and per-carrier health monitoring, so a route that degrades can be diagnosed and replaced rather than tolerated.

Frequently Asked Questions

What is a good MOS score?

The scale runs from 1 to 5, where 5 represents quality nobody would complain about and 1 is unusable. Scores in the low fours are generally considered good for voice over IP, the threes are noticeably imperfect but workable, and anything at two or below will generate complaints. Treat published thresholds as guidance and calibrate against what your own agents report.

How is MOS calculated on a live call?

Originally MOS came from panels of listeners rating recorded samples. On live traffic it is estimated instead, from the network measurements that predict perceived quality: packet loss, jitter, latency and the codec in use. The number that appears on a dashboard is therefore a model of how the call probably sounded, not a human opinion of it.

What causes a low MOS?

Almost always the network between the endpoints rather than the platform at either end. Packet loss produces dropouts, jitter produces choppiness, and excessive latency produces the talk-over effect where both parties start speaking at once. Overloaded shared connections and poor wireless are the usual sources on the agent side.

Does a codec change MOS?

Yes. A narrowband codec caps the best achievable quality no matter how clean the network is, while a wideband codec can sound noticeably better on the same connection. This is why comparing MOS across routes only means something when the codec is the same.

Why report MOS per call rather than as an average?

Because an average hides exactly the calls you need to find. A day with good average quality can contain a group of calls that were unusable, usually clustered on one agent, one route or one interval. Per-call scores let you find that cluster; an average tells you the day was fine.

Related terms: SIP trunk, BYOC, WebRTC softphone, wallboard and connect rate.