Call Quality Threshold Reference
Table of Contents
Latency (one way, mouth to ear) Packet Loss Jitter MOS Score How do they relate to each other? Notes for DiagnosisA quick reference for diagnosing MOS, latency, jitter, and packet loss on voice quality issues.
When a customer reports choppy audio, dropped words, or "bad call quality," the four measurements below are usually enough to confirm whether the network is the cause and how severe the impact is. Use this page alongside the raw numbers from the customer's router, SBC, or trace file.
Latency (one way, mouth to ear)
This is how long it takes a word to travel from the speaker's mouth to the listener's ear. Think of it like a walkie-talkie delay. A little delay is unnoticeable, but once it creeps past 300 ms, people start talking over each other because neither side realizes the other is still speaking. It is a timing problem, not a clarity problem.
| Range | Rating |
|---|---|
| Less than 150 ms | Excellent |
| 150 to 200 ms | Acceptable |
| 200 to 300 ms | Noticeable degradation |
| 300 ms and above | Poor, unusable |
Packet Loss
Every call is broken into hundreds of tiny data packets per second. Packet loss is the percentage of those packets that never arrive. A useful analogy: imagine someone reading you a sentence over a bad radio signal, and every so often, a word just disappears. Your brain fills in small gaps easily; that is what "concealment" does, but once too many words vanish, the sentence stops making sense.
| Range | Rating |
|---|---|
| 0% | Excellent |
| Up to 1% | Good, imperceptible with concealment |
| 1 to 3% | Noticeable degradation |
| 3 to 5% | Poor |
| 5% and above | Generally unusable |
Jitter
Jitter measures how consistently packets arrive, not whether they arrive at all. Picture a bucket brigade passing water buckets down a line. If everyone passes at a steady rhythm, the water flows smoothly. If some buckets arrive early and some arrive late, the flow gets uneven, and some buckets bunch up and spill. On a call, this unevenness is what causes that robotic or choppy sound, even when no data was actually lost.
| Range | Rating |
|---|---|
| Less than 20 ms | Excellent |
| 20 to 30 ms | Acceptable |
| 30 ms and above | Audible choppiness likely, even with jitter buffers |
MOS Score
MOS is the "report card" number. Instead of measuring a single technical factor, it is a 1-to-5 score intended to reflect what a real person would say if you asked them to rate the call. It is calculated from a combination of latency, jitter, and packet loss, so it is the fastest way to answer "how bad is this, really" without digging into three separate numbers.
| Range | Rating |
|---|---|
| 4.3 to 5.0 | Excellent (toll quality) |
| 4.0 to 4.3 | Good, typical enterprise Hosted Voice target |
| 3.5 to 4.0 | Fair, users start noticing |
| Below 3.5 | Poor, complaints likely |
How do they relate to each other?
| If you see... | It usually explains... |
|---|---|
| High latency alone | Awkward pauses, people talking over each other |
| High packet loss | Words or syllables cut out, garbled speech |
| High jitter | Choppy or robotic-sounding audio |
| Low MOS | A summary that something above is off, check the other three to find which one |
Notes for Diagnosis
- Burst loss versus scattered loss. Consecutive lost packets degrade MOS far more than the same percentage spread evenly across a call. Ask the customer, or check the trace, for whether the loss was continuous or scattered before rating severity.
- Round-trip latency above 300 ms. This causes talk-over and interruptions regardless of what the MOS math says, because it breaks natural conversational turn-taking. Call this out separately from the MOS rating whenever it is reported.
- Codec matters. G.711 tolerates loss worse than Opus or G.729 with good packet loss concealment (PLC). Confirm which codec is in use on the call before comparing the numbers against these thresholds.