Talk to Michelle.
Three engines, two voices.
Three voice AI engines, side by side. Talk to Michelle on each, feel how she responds, and pick the model that fits the way your business actually runs.
Click your microphone to allow access. Sessions cap at 3 minutes.
Michelle on GPT 4.1 (default)
Best for cheapest reliable text-llm.
Michelle on GPT 4.1 Fast
Best for same model on priority routing.
Michelle on GPT Realtime 2.0
Best for natural, fluent conversation.
Web demo only. Add roughly 200 ms for the carrier path on a real phone call. The relative gap between engines stays the same.
Side-by-side capabilities
Quick scan of what each engine actually does well.
| Capability | GPT 4.1 | GPT 4.1 Fast | Realtime 2.0 |
|---|---|---|---|
| Sticks to a structured script | |||
| Cheapest per minute on a budget call | |||
| True speech-to-speech engine | |||
| Lowest first-audio latency | |||
| Natural turn-taking | |||
| Handles interruptions cleanly | |||
| Custom NZ and AU accent voices | |||
| 128K context for long calls |
What kind of calls is each model best at?
Pick the right model for the job, not the loudest one in the room.
Reach for it on scripted, high-volume calls
- ·Cold outbound dialling a dormant database (10,000 contacts at the lowest cost per minute).
- ·Appointment reminders the day before, where the script barely changes.
- ·Lead qualification with a fixed five-question script before a human takes over.
Reach for it when 200 ms of dead air costs you the call
- ·Cold outbound at scale where the same script needs a snappier delivery.
- ·When the caller's answers run long and you need faster speech to text turnaround. Best fit when the brand wants a more premium quality sounding agent.
- ·Anything where you have run the default 4.1 and seen too many hangups.
Reach for it when the conversation has to feel natural
- ·Lowest latency model on the page. Snappy replies that feel instant.
- ·Depth and nuance in how it answers, not just speed.
- ·Warm vendor or client callbacks where rapport carries the call.
Hundreds of voices.
Any persona, any business.
Tell us your industry, your accent, your call flow, and your CRM. We build the voice agent that fits, with the voice that fits the brand. NZ, AU, UK, US accents and every persona between.
Frequently asked
- Why do the two Michelles sound different?
- GPT Realtime forces you onto OpenAI’s voice catalogue. We can run our custom Michelle voice with GPT 4.1 because it’s text plus our own TTS layer. We can’t do that with the Realtime engine yet. We pinned the script so the AI behaviour is what you compare, not the voice.
- Are these real production agents?
- No. These are demo agents with a brief Michelle persona. Production Waboom AI agents have full booking, transfer, and post-call analysis wired in.
- What does the latency number mean?
- Time from when Michelle decides you stopped speaking to the first byte of audio coming back to your browser. Real phone calls add roughly 200 ms over the carrier path.
- Want one on your real number?
- Book a call and we will dial you with whichever engine you want to feel.