Objective
What you're building
Replace the default voice from Lesson 1 with a real, chosen ElevenLabs voice — tuned, not just picked at random — and prove the difference by calling both back to back.
Voice settings aren't decoration — they change how natural it sounds
{
"voice": {
"provider": "11labs",
"voiceId": "your-chosen-voice-id",
"stability": 0.5,
"similarityBoost": 0.75,
"style": 0.3
}
}
stability controls how consistent the voice sounds across a conversation (higher = more consistent but can sound flatter; lower = more expressive but can sound erratic). similarityBoost controls how closely the output matches the original voice sample (relevant mainly for cloned voices). style adds expressiveness/emotion at some cost to stability. There's no universally correct setting -- a receptionist wants consistency and clarity over dramatic expressiveness, which is a real, deliberate choice among these tradeoffs, not a default to accept blindly.
More expressive settings can add latency
Higher style values and more complex voice models can take marginally longer to generate per response -- in a domain where Lesson 1 already established that every hop's latency matters, a slightly more "flat but fast" voice setting can genuinely be the right tradeoff for a business phone line, even though it sounds less impressive in isolated testing.
Compare, don't just pick once and move on
You pick a voice that sounds great in ElevenLabs' web preview (reading a sample script) but sounds slightly off during an actual Vapi phone call with real conversational back-and-forth. What does this tell you?
A voice preview reads a fixed script under ideal conditions; a live call generates responses to your actual words, with the phone codec's compression and Vapi's real-time pipeline in the loop. These are different enough tests that "sounded great in preview" doesn't guarantee "sounds great on a real call" -- which is exactly why this course keeps insisting on the actual-phone-call test over dashboard/preview checks.
What you're building today
voice_config.json: a specifically chosen ElevenLabs voice (not the default) with tuned stability/similarityBoost/style values, verified by making 2 real calls back to back (default voice vs. your chosen one) and noting the actual, audible difference.