Monitoring call quality on a carrier

Call quality on a Carrier is measured by four numbers: MOS, jitter, packet loss, and latency. Watch them per route, alert when they cross a threshold, and you will catch a degrading carrier before agents start complaining and conversions start dropping.

The four metrics that matter

• MOS (Mean Opinion Score): a 1-to-5 summary of perceived audio quality. MOS (mean opinion score) above 4.0 is good, 3.5 to 4.0 is acceptable, below 3.5 is where agents notice. It is derived from the other three.
• Jitter: variation in packet arrival time. Jitter under 30 ms is fine; high jitter makes audio choppy because the buffer cannot smooth it out.
• Packet loss: the share of RTP packets that never arrive. Packet loss over 1 percent starts to be audible as clipped or robotic speech; over 3 percent is bad.
• Latency: one-way delay between the two ends. Latency under 150 ms feels natural; past 300 ms people start talking over each other.

Where degradation comes from

Bad numbers usually trace to one of a few places. The carrier itself may be overloaded or routing your calls through a congested path. Your own uplink may be saturated — running too many concurrent calls on a thin connection starves RTP of bandwidth. A long geographic path adds unavoidable latency. And the wrong Codec choice matters: a compressed codec like G.729 codec saves bandwidth but is less forgiving of loss than uncompressed G.711 codec.

From metric to action

flowchart TD
  A[Collect MOS jitter loss latency] --> B{Within thresholds?}
  B -->|yes| C[Keep dialing]
  B -->|no| D{Which metric?}
  D -->|high jitter/loss| E[Check uplink + concurrency]
  D -->|high latency| F[Check route / carrier path]
  D -->|MOS keeps dropping| G[Alert + fail over to backup carrier]
  G --> H[Switch route, re-measure]

The loop is simple: measure, compare against a threshold, and if a route is consistently bad, move traffic to a backup carrier and re-measure. The key word is consistently — one bad call is noise, a route that sits below MOS 3.5 for an hour is a problem.

Setting up alerting

1. Pick thresholds per metric: alert below MOS 3.5, above 1 percent loss, above 30 ms jitter, or above 200 ms latency. Tune to your tolerance.
2. Sample continuously, not once. Quality drifts with time of day and call volume, so you want a trend, not a single reading.
3. Keep a second carrier configured and ready. The fastest fix for a degrading route is to stop sending traffic to it, which only works if the alternative is already wired up.
4. Correlate quality dips with drop rate and abandons — bad audio quietly inflates both before anyone files a ticket.

Quality monitoring only pays off when you can act on it, which means having more than one trunk to switch between. For how to wire that second carrier in, see our carrier integration guide, and for running several trunks across servers, see VICIdial server trunks.

A dialer that runs on a fixed public IP with predictable bandwidth removes one whole variable from the quality equation. See our plans — you bring your carriers, we give you a stable box to measure them from in under 40 seconds.