Nightingale — clip voice audition

gpt-4o-mini-tts · 11 voices · 8 kHz · 2026-08-07

This is the recording-consent notice — 17 seconds, the first thing every caller hears, and on a short call most of the audio they hear at all. A voice that works here works everywhere; one that grates here defines the demo. Every take below is rendered through the same path that ships, so what you pick is exactly what deploys.

Thanks for calling Northgate Family Dental. Our office is closed right now, so you're speaking with an automated assistant. This call is recorded so our team can follow up with you in the morning. If you'd rather not be recorded, please hang up and call back during business hours.

What's live right now

Rendered with the old direction: “a recorded announcement on a telephone line… do not sound cheerful.” That instruction is why it lands flat — the clip was doing what it was told.

ash live now 18.3s

The candidates — new delivery direction

Same sentence, same pipeline, re-directed: “a real person at the front desk who recorded the after-hours greeting themselves — warm, natural, unhurried, never flat or robotic.” The other three clips get their own directions; the 911 line is written for calm authority rather than warmth, because a caller who mistakes it for a recording may not act on it.

alloy 15.9s
ash 17.0s
ballad 18.8s
coral 17.9s
echo 15.9s
fable 19.0s
nova 15.7s
onyx 16.1s
sage 20.9s
shimmer 15.7s
verse 16.9s

How to pick

  1. Worst speakers you own. Laptop or phone, not headphones.
  2. It has to survive 17 seconds. Judge the back half — anything can hold up for one sentence. Fatigue is the failure mode here, not first impression.
  3. Keep it clearly distinct from marin, the agent's voice. When the system interrupts to say something the practice is accountable for, the caller should be able to hear that something changed.
  4. Then imagine the 911 line in it. The same voice delivers “hang up now and call 9-1-1.” Warm is right for the greeting; it must still be able to carry weight.