Latency
What is Latency?
Latency is the time delay between a trigger and its observable effect — in networking, the time a packet takes to travel from sender to receiver; in voice AI, the pause between a caller finishing speaking and the agent's reply; and in APIs, the time between a request being sent and a response returning. It's usually measured in milliseconds and tracked alongside jitter and packet loss as one of the core metrics determining call and API quality.
More detail
Latency comes from the network path itself (distance, hops, queuing) and from the endpoints (processing time, buffer delays). A voice call with 50 ms of mouth-to-ear latency feels natural; at 150-200 ms it starts to feel laggy, and at 400 ms+ the conversation breaks down into turn-taking struggles.
In telecom debugging, latency is one of three QoS metrics together with jitter (variation in delay) and packet loss (missing packets). A high-speed low-jitter link with moderate latency is still preferred over a low-latency link suffering loss or erratic timing.
Inside a voice AI pipeline, latency decomposes per stage: network round-trip, speech-to-text streaming, model reasoning, and text-to-speech generation. A production pipeline streams each stage to avoid summing them sequentially. Post-dial delay is a special case of latency — the wait between finishing dialing and hearing a ring.
Latency is distinct from throughput: throughput measures how much data flows per second; latency measures how long a single unit takes. A satellite link may have great throughput while suffering hundreds of milliseconds of latency.
Frequently asked
- Is latency the same as lag?
- Lag is the colloquial term for the slowdown a person perceives; latency is the measurable time delay behind it. The two are often discussed together, but they're not synonyms — latency is a number in milliseconds, lag is the experience that number produces.
- What latency is acceptable for a phone call?
- Under ~50 ms mouth-to-ear a call feels natural; 150-200 ms becomes noticeably laggy but usable; above ~400 ms callers start interrupting each other. The acceptable threshold depends on the medium — a satellite link tolerates high latency where a wired voice call would not.
- How does latency interact with jitter and packet loss?
- Latency is one QoS metric alongside jitter (variation in delay) and packet loss (dropped packets). A low-latency path suffering consistent jitter or loss is often worse than a slightly slower path with stable timing, which is why call-quality diagnostics report all three together.
- Does latency only matter for voice?
- No — every real-time system is latency-sensitive: an API response time, a webhook round trip, a WebSocket message delay, a head-of-queue flush. The same concept applies, just measured against the appropriate unit (a call, a packet, an HTTP request).
See also
Build it on Orbit
Voice, messaging, email, video, and AI agents on one platform and one pay-as-you-go bill. Start free — no credit card required.