What changed in how Vozi sounds
As of 5 October 2026, Vozi speaks with a new voice model. Ring the demo line and you'll hear the difference in the first sentence.
It's the biggest change to how Vozi sounds since we launched. The receptionist still knows the same things, books into the same calendars and follows the same call rules. What changed is the voice itself: the accent, the emotion behind the words, the way it says names, and how soon it starts answering.
Most of the work went into the things callers notice without being able to say why. A voice can be perfectly clear and still sound like a recorded message. That's the gap we wanted to close.
An Australian accent that lasts the whole call
The new voice holds its Australian accent from "G'day" to "see you Thursday", including the long sentences where the old one used to drift.
The old voice had a good accent on short lines. Give it a long sentence, or a string of numbers, and it could wander somewhere mid-Pacific. Reading back an address in Croydon North with a unit number and a postcode was where you'd hear it most.
The new model keeps the same voice and accent through long passages. Vowels stay Australian. The rhythm of a mobile number read back in pairs sounds like a person at a front desk would say it, with the small pause before the last three digits.
It matters more than it sounds. Callers relax when the voice on the other end sounds local; we saw that in the work we did on Aussie accents, and it's the thing customers comment on first.
Why it sounds less like someone reading a script
The new voice reads the conversation before it speaks, so its tone follows what is actually happening on the call.
The old voice was good at a friendly, even read. It was weaker at the moments that matter most on a small business line: a caller whose hot water system died at 9pm, someone ringing to cancel, an older patient who needs things said slowly and clearly. It said everything the same way.
Now the delivery shifts with context. An apology sounds like one. Confirming a Thursday 2pm booking sounds brisk and certain. We can also give the voice direction for particular moments (slow down when reading back a mobile number, for example), and it follows those directions far more reliably than the last one did. It handles the awkward parts better too, like putting someone on hold while it checks a calendar, or bringing the temperature down when a caller is already annoyed.
What surprised me in testing was the small stuff. A soft laugh when a caller makes a joke. A breath before a longer sentence. None of that shows up in a transcript, and it changes how the whole call feels.
Getting Woolloomooloo right first time
Custom pronunciations now hold. When we tell Vozi how to say a suburb, a street or a surname, it says it that way on every call.
This was one of the most common fixes customers asked for. Australia is full of place names that trip up any computer voice: Woolloomooloo, Mullumbimby, Wagga Wagga, Toowoomba, Kurri Kurri. Vozi has always understood them well when a caller says them; saying them back properly was the weaker half.
Then there are surnames. A dental clinic in Box Hill doesn't want its receptionist mangling Dr Nguyen's name forty times a day.
We could set pronunciations on the old voice, but they didn't always stick, especially halfway through a long sentence. The new one takes a phonetic spelling and keeps to it. If there's a word Vozi gets wrong for your business, send it to us and we'll fix it.
How quickly it answers now
Most replies now start 0.8 to 1.2 seconds after the caller stops talking. Across our first 100,000 calls the median was 1.34 seconds.
That gap covers everything: hearing the caller out, working out the answer and starting to say it (we explain that whole chain in how a voice AI agent works). The new voice trims its share by starting to speak while the rest of the answer is still being put together, so a long reply no longer waits for its own last word.
Dead air kills phone calls. Once a pause stretches past a couple of seconds, callers start saying "hello?" and talking over the reply.
I don't think it should be instant, though. People pause between turns, and a reply with no gap at all sounds like a machine. Around a second is where it feels like a conversation. You can compare it with the old numbers in what 100,000 phone calls taught us.
What happens when a caller switches to Mandarin
The new voice speaks more than 90 languages. It does that in the same voice, with a native accent in each one.
Mandarin is the most common language other than English spoken at home in Australia, according to the 2021 Census, followed by Arabic, Vietnamese, Cantonese and Punjabi. For a medical centre in Glen Waverley or a property manager in Cabramatta, a real share of callers would simply rather speak another language. Mandarin and Cantonese got a clear quality bump in this release, which matters more here than in most countries.
I want to be careful here. Every Vozi call flow is written and tested in English. The voice can switch languages mid-call, but we haven't tested booking rules or transfer logic in other languages with the same care. If multilingual calls matter to your business, talk to me before you rely on it.
The release on one page
- Released
- 5 October 2026. Live on the demo line, so anyone can hear it.
- Accent
- Australian, held consistently through long sentences, numbers and addresses.
- Tone
- Follows the context of the call, and takes direction for specific moments such as reading back numbers slowly.
- Pronunciation
- Phonetic spellings for suburbs, streets, product names and surnames, applied on every call.
- Reply time
- Typically 0.8 to 1.2 seconds from the caller finishing to Vozi starting to answer.
- Languages
- More than 90, in one voice, with improved Mandarin and Cantonese.
What's missing: a full month of live-call data on the new voice. Sentiment and transfer rates will go in a follow-up once there's enough of it to mean something.
What the new voice doesn't fix
If I were a business owner reading this, these are the things I'd want to know.
A better voice doesn't make a wrong answer right. If Vozi has been given the wrong Saturday hours, it will now tell callers the wrong hours more warmly. The information it's given still decides most of how a call goes.
It's still an AI, and most of our customers have it say so. We wrote about whether to tell callers they're talking to an AI last year; a voice this natural makes that disclosure matter more to me.
The phone network hasn't changed. A caller on one bar of signal somewhere along the Hume Highway will still hear a delay that has nothing to do with the voice.
What's next
Next month I'll publish call data on the new voice, measured the same way as the 100,000 calls report, so you can put the two side by side. I'm most interested in whether the negative-sentiment share drops now that the voice can actually sound sorry.
The quickest way to judge all this is to hear it. Get Vozi to ring you, ask it something awkward, and tell me how it went. My LinkedIn is just below and I read every message.
Ring the new voice yourself.
Enter your mobile and Vozi calls you within a minute. Or book a 15-minute setup call and we'll have it answering your phone by tomorrow morning.
Nick Davenport
Founder of Vozi, an Australian voice AI receptionist used by small businesses in trades, legal, health and property. Nick works directly with customers on their call flows and reviews call data every month.
Nick on LinkedIn