Jitter, loss, and what the ear tolerates

“The call quality is bad” can be translated into measurements, and the translation is reliable enough to diagnose from a description.

What each condition sounds like

Latency — people talk over each other and pause awkwardly. Audio is clear. Under about 150 ms one way, unnoticeable. Beyond 300 ms, conversation becomes turn-taking.

Jitter — variation in arrival time. The jitter buffer absorbs it by holding packets and playing them evenly, which costs delay. Beyond the buffer’s capacity you get choppy or robotic audio.

Loss — gaps. Modern codecs conceal isolated losses well; a few tenths of a percent is inaudible. Consecutive losses are not concealable and produce clearly dropped words. Bursty loss is far worse than the same percentage spread evenly.

Echo — hearing yourself. Usually an analogue interface or a device’s acoustic path, not the network. It becomes audible only when delay is high enough to separate it from your own speech, which is why it appears when a path gets longer.

Measuring rather than guessing

RTP carries the data; RTCP carries the quality reports. Phones and gateways record jitter, loss and round-trip time per call. Those statistics, per call, are the evidence.

show call active voice brief
show voice call status
# in Wireshark: Telephony > RTP > Stream Analysis

Targets worth holding

One-way delay under 150 ms where the physics allow. Jitter under 30 ms. Loss under 1%, and ideally well under. MOS above 4.0.

The pattern that identifies the cause

Bad calls only at busy times means congestion, and QoS is the answer. Bad calls to one destination means a path problem, not a capacity one. Bad calls from one handset means the handset or its cable. Bad calls at random with no pattern usually means jitter from a device doing something in software that should be in hardware.

Ask when it happens before asking what is wrong. The timing is usually the diagnosis.