Call-quality measurements

We measure call quality by placing real phone calls and reading what the call itself reports: how long sound takes to travel, how uneven its timing is, and how much of it goes missing. Those three numbers are turned into one score from 1 to 5 using ITU-T G.107, the international standard for the job, so anyone can check the arithmetic.

Every result below is published with the individual calls behind it, including calls that failed. So far we have measured one network. The providers we rank need test accounts before we can measure them the same way, and until then their measured columns stay empty rather than guessed.

Latest result: Comexcel

Measured September 23, 2026, 9:50 PM–9:58 PM Eastern

4.38Excellentcall-quality score (MOS) · 4.0 and above sounds professional · ~4.4 is the ceiling

9 of 10 test calls measured, 30 seconds each, codec G711.

Delay, one way: 27.0 msComfortable under 150 msHow long your voice takes to reach the other end. Well under the point where people start talking over each other.
Jitter: 6.8 msFine under 30 msHow uneven the timing is. Within range, and partly caused by the Wi-Fi this was measured over.
Lost audio: 0.0%Fine under 1%Sound that never arrived. None was lost on any measured call.

Where and how this was measured

Connection: A laptop on Wi-Fi, on a Comcast broadband connection in Broward County, Florida, routed through Miami
Link: Wi-Fi
What was called: Comexcel's PBX echo test (*43). The PBX runs in Amazon Web Services' us-east-1 region in Northern Virginia, so this measures the full round trip from a Broward County office to the phone system and back.
Scoring: ITU-T G.107 E-model applied to RTCP statistics from real calls

What this does and does not tell you

It shows how calls between this connection and Comexcel's phone system sounded when it was measured. The round trip covers the whole internet path in between, in both directions.
It does not measure the far end of a real call to a customer, which depends on their phone company too, and it cannot say how the same system sounds from your office.
It was taken over Wi-Fi, which adds timing variation of its own, so on a cable the jitter figure would be expected to be lower. That is also why uneven timing is usually a local problem rather than the provider's.
One evening and one connection is a small sample. We will repeat it monthly and from a wired connection, and publish every run here, including the unflattering ones.

Every call in this run

Times are Eastern. Round trip is what the call reports; the score uses half of it as the one-way delay.

Individual test calls in the latest measurement run
Call	Started	Round trip	Jitter	Lost	Audio packets
1	9:50 PM	48.2 ms	6.5 ms	0.0%	1,100
2	9:51 PM	53.7 ms	7.0 ms	0.0%	1,100
3	9:52 PM	54.6 ms	6.9 ms	0.0%	1,100
4	9:53 PM	53.8 ms	6.8 ms	0.0%	1,100
5	9:53 PM	60.4 ms	6.8 ms	0.0%	1,100
6	9:54 PM	42.9 ms	6.8 ms	0.0%	1,100
7	9:55 PM	80.4 ms	6.7 ms	0.0%	1,100
8	9:56 PM	69.2 ms	6.9 ms	0.0%	1,100
9	Not used in the averages: answered, then ended by the far end after 6.4 seconds with no audio received.
10	9:57 PM	22.3 ms	6.6 ms	0.0%	1,100

Link type recorded after the run by checking the machine's default network interface (en0, Wi-Fi) immediately afterwards; the runner now records it automatically.
Wi-Fi adds delay variation of its own, so the jitter here is an upper bound for this network path. A wired connection would be expected to show less.
Call 9 reasons were re-derived from its raw log with the current parser after the run; the averages are unchanged.

Raw result file

{
  "kind": "sip-call",
  "target": "comexcel",
  "metro": null,
  "date": "2026-09-24",
  "measuredAt": "2026-09-24T01:58:17.396Z",
  "vantage": "broward-comcast",
  "vantageDescription": "A laptop on Wi-Fi, on a Comcast broadband connection in Broward County, Florida, routed through Miami",
  "testDescription": "Comexcel's PBX echo test (*43). The PBX runs in Amazon Web Services' us-east-1 region in Northern Virginia, so this measures the full round trip from a Broward County office to the phone system and back.",
  "callsAttempted": 10,
  "calls": 9,
  "rejected": [
    {
      "call": 9,
      "startedAt": "2026-09-24T01:56:55.282Z",
      "reason": "answered, then ended by the far end after 6.4 seconds with no audio received"
    }
  ],
  "callDurationSec": 30,
  "codec": "g711",
  "rxPacketsAvg": 1100,
  "owdMs": 26.98,
  "jitterMs": 6.77,
  "lossPct": 0,
  "rFactor": 92,
  "mos": 4.38,
  "verdict": "Excellent",
  "method": "ITU-T G.107 E-model applied to RTCP statistics from real calls",
  "samples": [
    {
      "call": 1,
      "startedAt": "2026-09-24T01:50:46.799Z",
      "rttMs": 48.16,
      "jitterMs": 6.531,
      "lossPct": 0,
      "rxPackets": 1100,
      "codec": "PCMU"
    },
    {
      "call": 2,
      "startedAt": "2026-09-24T01:51:32.867Z",
      "rttMs": 53.738,
      "jitterMs": 6.956,
      "lossPct": 0,
      "rxPackets": 1100,
      "codec": "PCMU"
    },
    {
      "call": 3,
      "startedAt": "2026-09-24T01:52:18.898Z",
      "rttMs": 54.647,
      "jitterMs": 6.891,
      "lossPct": 0,
      "rxPackets": 1100,
      "codec": "PCMU"
    },
    {
      "call": 4,
      "startedAt": "2026-09-24T01:53:05.001Z",
      "rttMs": 53.771,
      "jitterMs": 6.81,
      "lossPct": 0,
      "rxPackets": 1100,
      "codec": "PCMU"
    },
    {
      "call": 5,
      "startedAt": "2026-09-24T01:53:51.035Z",
      "rttMs": 60.44,
      "jitterMs": 6.818,
      "lossPct": 0,
      "rxPackets": 1100,
      "codec": "PCMU"
    },
    {
      "call": 6,
      "startedAt": "2026-09-24T01:54:37.091Z",
      "rttMs": 42.86,
      "jitterMs": 6.763,
      "lossPct": 0,
      "rxPackets": 1100,
      "codec": "PCMU"
    },
    {
      "call": 7,
      "startedAt": "2026-09-24T01:55:23.159Z",
      "rttMs": 80.361,
      "jitterMs": 6.663,
      "lossPct": 0,
      "rxPackets": 1100,
      "codec": "PCMU"
    },
    {
      "call": 8,
      "startedAt": "2026-09-24T01:56:09.222Z",
      "rttMs": 69.239,
      "jitterMs": 6.887,
      "lossPct": 0,
      "rxPackets": 1100,
      "codec": "PCMU"
    },
    {
      "call": 10,
      "startedAt": "2026-09-24T01:57:41.326Z",
      "rttMs": 22.338,
      "jitterMs": 6.601,
      "lossPct": 0,
      "rxPackets": 1100,
      "codec": "PCMU"
    }
  ],
  "vantageLink": "wifi",
  "notes": [
    "Link type recorded after the run by checking the machine's default network interface (en0, Wi-Fi) immediately afterwards; the runner now records it automatically.",
    "Wi-Fi adds delay variation of its own, so the jitter here is an upper bound for this network path. A wired connection would be expected to show less.",
    "Call 9 reasons were re-derived from its raw log with the current parser after the run; the averages are unchanged."
  ]
}

Earlier runs

Comexcel, August 31, 2026: score 4.40 (Excellent) from 3 calls. Recorded before we kept individual calls and the connection type, so it is listed for the record rather than shown in detail.

What happens next

The providers we rank will be measured the same way once we have test accounts with each of them, from the same connection and in the same window, so the numbers can be compared fairly. Until then the measured columns on our comparison pages stay empty. The full method, and how we make money, is on the methodology page. You can test your own connection in the meantime with our VoIP speed test.