Jitter
Jitter is the variation in how long audio packets take to travel across a network, causing packets from the same voice stream to arrive at uneven intervals instead of a steady rhythm.
Internet voice systems split speech into small packets and send them independently. The packets may take different routes or encounter temporary congestion, so their arrival times can vary even when none are lost. A receiving system usually holds incoming packets briefly in a jitter buffer, then plays them in the correct order at a regular pace.
That buffer creates a tradeoff. A larger buffer can smooth out greater timing variation, but it also adds delay. A smaller buffer keeps the conversation more immediate, but a late packet may miss its playback window and become unusable. Adaptive jitter buffers change their holding time as network conditions change.
For an AI phone call, jitter matters because the caller and agent both depend on a continuous audio stream. Excessive variation can produce clipped syllables, short gaps, robotic-sounding speech, or apparent interruptions. It can also make turn boundaries harder to interpret: a delayed fragment may arrive after the system has started treating the caller's turn as complete.
Jitter is different from latency. Latency is the total time audio takes to move from speaker to listener, while jitter describes how inconsistent that travel time is from packet to packet. A call can have noticeable latency but little jitter, or low average latency with disruptive spikes in jitter. It is also distinct from packet loss, although packets that arrive too late for playback are often treated much like lost packets.
When troubleshooting, teams should examine the network path, congestion, wireless stability, packet prioritization, and jitter-buffer behavior together. Listening to a recording alone may not reveal whether the problem occurred before transmission, on the network, or during playback. The useful goal is not zero timing variation, which is unrealistic, but stable audio delivery without adding enough buffering to make conversation feel sluggish.