A pause is the only thing that gives an AI away.
So we measure it. Below is the turn latency of every Nutaan voice engine and delivery style — the gap between a caller finishing their sentence and hearing your agent start — measured on real phone audio, not in a lab.
Voice engines
The engine decides how a call is put together. The three speech-to-speech engines answer in well under a second because there are no stages to hand off between; the Standard pipeline is slower per turn but the most flexible and the cheapest to run at volume.
Elite delivery styles
On the Elite engine you also choose how the voice performs. Expression and speed pull against each other — these are the four points on that curve.
How we measure
Turn latency is measured end to end: from the moment the caller stops speaking to the first audio frame of the reply leaving our servers. Every run uses the same voice, the same sentence and telephony-grade μ-law audio at 8 kHz, so the numbers compare like for like.
Figures are typical values, rounded. Real calls vary with network conditions, how long the reply is, and how far the caller is from the nearest region.
Where a range is shown, the bar spans the fastest and slowest turns we measured, and the tick marks where a turn typically lands. Expressive has the widest spread of anything here — 140 ms to 2.5 s. It buys the most emotion, so it suits a persona that has to feel alive and not a queue where every second counts.
Speed is not the only axis. The fastest engine is English-only, the most expressive voice is not the quickest, and the cheapest engine is the slowest. The right choice is the one that fits the call — which is why every option is switchable per agent.
Hear it for yourself.
Numbers only go so far — take a live call with an agent in your language and judge the pause yourself.