Skip to content

glossary

verified July 2026

Time to first audio (TTFA)

Time to first audio is the delay between handing text to a TTS system and receiving the first byte of playable audio. For a voice agent it is the number that decides whether the agent answers or hesitates.

01

Where the milliseconds land

The scale below is one second — the rough ceiling of a caller’s patience. Everything about a TTFA claim is where it sits on this line, and how far its p95 drifts from its p50.

Fig. — One second of caller patience, to scale

107 / 108 ms

p50 / p95, single stream, production API

200 ms

human turn-taking rhythm

1,000 ms

callers start interrupting

107 / 108 ms

p50 / p95, single stream, production API

200 ms

human turn-taking rhythm

1,000 ms

callers start interrupting

The gap between p50 and p95 is the honesty of a latency claim — ours is one millisecond on a single stream, which is why the two marks sit on top of each other.

02

TTFA vs model latency vs TTFB

Model latency counts only the neural network’s forward pass — no API gateway, no queue, no network. Time to first byte (TTFB) is a web metric that counts any response byte, including headers. TTFA counts what the caller experiences: request sent, first audible sample back. A vendor quoting model-only latency can be honest and still describe a number you will never see in production.

03

How to measure it honestly

  • Measure server time on the production endpoint, not a staging box or a co-located benchmark.
  • Use call-shaped load — short conversational utterances, not paragraph batches.
  • Report percentile pairs: a p50 without a p95 hides the tail your callers live in.
  • State what the figure includes. Ours: 107 ms p50, 108 ms p95, server time to first audio on the production API.

See also

Related sheets.

Every term on this page is measurable on a live call — watch the readout while your own script synthesizes.

See the measured figures live