Low lets her voice move with the sentence. Too low and she drifts between turns.
How tightly she holds the chosen voice's timbre.
Adds performance, and 20–50 ms of latency per utterance. Zero is the safe default.
1.08 trims ~8% off every utterance. At 1.0 a 30-word opener took ~18s — too slow for a phone.
How much her wording changes between callers. High wanders off script.
Sets instructed reply length and how much of the campaign's knowledge reaches the brain. Not model reasoning — that would make every turn slow.
Writes these onto the campaign, so Script Studio and Run Campaign use them from the next call.
Pick a campaign to test against.
The test runs the real stack — same STT, brain, LLM and voice the phone uses. It reads the campaign as saved, so press Apply before testing a change you just made with the sliders.