| | Episode #10 The VoiceAISpace Newsletter | |
| | | 3 × News
2 × Products
1 × Spotlight
Space Updates | |
|
|---|
| | This week, a Singapore hospital is giving nurses some of their typing time back, New Orleans has discovered that "AI answering 911 calls" is a phrase you probably want to explain very, very carefully, and the humble family calendar has apparently decided it would rather be a voice agent.
Elsewhere, AI gets its own physical button, an English teacher moves into Minecraft, and we meet a founder who started building Voice AI early enough to survive the pre-LLM years and stick around for the "oh... now this actually works" moment.
Then came the global moment: nine cities, one day. From Melbourne to San Francisco, the Voice AI community took the conversation to real-life tables; same geeky energy, just with better snacks. Builders swapped ideas, conversations ran long, and no one seemed in much of a hurry to leave. Turns out, Voice AI travels pretty well.
Hospitals, 911 calls, Minecraft and a tiny AI button. Not a bad week in Voice AI.
Quite a nice spread, actually. | |
03 | News The Good, The Bad & The Funky |
| | Singapore's National University Hospital (NUH) has found a fairly sensible use for Voice AI: letting nurses spend a little less of their day being extremely expensive typists. Instead of finishing an encounter and then sitting down to document everything that just happened, nurses can speak naturally while tools including NUH's in-house MediVoice and Singapore Open Government Products' Scribe turn the conversation into structured clinical notes. The systems are being worked into real nursing workflows, it supports multiple languages and dialects, and here's the number that caught our attention early; implementation data suggests they can save up to 16 minutes per encounter in selected use cases. The AI doesn't get the final say, nurses still check the notes before they go anywhere important. And those 16 minutes? Multiply them across a busy hospital and suddenly that's a lot less screen time, and a lot more actual nursing.
Why we like it:
This is the sort of Voice AI deployment we enjoy seeing because the technology isn't desperately trying to become the star of the show. Nobody needs an AI nurse pretending to be a nurse. They need the actual nurse to spend less time typing what already happened.
Speech is simply the easier input here. If that means less documentation and a little more time with patients, good. Voice AI has done its job.
| |
| | AI is answering 911 calls. Well... sort of.A story spread around the US suggesting that New Orleans was using AI to answer 911 calls instead of human dispatchers. Understandably, people were not thrilled. There are few sentences capable of producing public anxiety faster than "the emergency line is being handled by a bot." The reality is considerably narrower. The Orleans Parish Communication District does use an AI system, but Axios reports that it steps in only under specific conditions: someone is calling from roughly 200 metres of an already-reported vehicle crash, the call appears to relate to that incident, and human dispatchers are already occupied. The AI basically asks whether the person is calling about the crash. Anything outside that scenario goes back to a human. So no, New Orleans has not handed 911 over to an autonomous AI dispatcher. But the original framing travelled far enough that officials started receiving concerned calls before the story was corrected. Which creates a slightly unusual Voice AI problem: the technology may be behaving more carefully than the story being told about it.
Why we're flagging it:
911 is about as high-trust as a voice interface gets. People need to know when they're speaking to AI, what that AI is allowed to do and, perhaps most importantly, what it isn't allowed to do.
The actual New Orleans implementation sounds deliberately constrained. The public reaction is still useful, though. Deploy Voice AI somewhere this sensitive and product communication becomes part of the safety system. You don't really get to be vague.
| |
| | Remember when the family calendar was a sheet of paper stuck to the fridge, squeezed somewhere between a takeaway menu and a school reminder? Cozyla clearly thought that was far too peaceful. Its new Calendar Plus 3 and Calendar Neo displays come with a conversational voice agent, so families can add events, organise routines, assign chores and generally keep household chaos under control by talking to the screen. Plus 3 takes on more of the planning; Neo keeps things a little simpler. Then there's Routie, a camera-free bedside companion for children. It can handle morning and bedtime routines, sync with the family calendar, and even read stories in a cloned version of a parent's voice. And yes, there's a physical microphone-off switch, reassuring for something living next to your child's pillow. These are Cozyla's own product claims rather than independent testing, but still: a calendar that takes voice commands and a bedtime companion that sounds like Mum or Dad? Yep. Funky enough for us.
Why it's funky:
Voice agents spend a lot of time pretending to be receptionists, salespeople, and customer-support reps. Cozyla brings that idea into everyday family life, helping with calendars, chores, routines, and the usual household juggling.
We actually think that shift is interesting. Voice AI may become mainstream less through a giant "VOICE AI" app and more by quietly turning up inside objects people already understand: the car, television, bedside device, kitchen screen... family calendar.
The fridge magnet had a decent run.
| |
Products Voice AI putting in the work | 02 |
| |  | #1 |
Button Computer is almost aggressively simple.
You clip a tiny computer to your clothing.
You press the physical button.
You talk.
It answers.
That's basically the product.
Button is a wearable computer designed specifically around Voice AI rather than adapting a phone, watch, or laptop interface that existed long before LLMs. The company says responses can begin in around half a second, and the device can answer through its built-in speaker or Bluetooth headphones.
The part we find particularly interesting is what it doesn't do.
The microphone isn't sitting there continuously listening to your life. It activates when you press the button. No attempt to constantly predict what you need. No notifications trying to drag you back into another screen.
Press. Ask. Get an answer. Move on.
Button is currently available for $179 on preorder, with the first U.S. shipments planned for late 2026. It also lets users bring their own API key instead of requiring the company's AI subscription.
The founders previously worked together at Apple on Vision Pro, which makes the hardware-first thinking slightly less surprising. Why we like it: most Voice AI products still live inside hardware designed for something else. Button asks a more interesting question: what if Voice AI deserves its own computer? We also like the intentionality of press-to-talk. There is an increasingly weird assumption in AI hardware that "useful" means "always listening." Maybe the better interface is just… a button. Technology you invoke when you want it and forget about when you don't. Quite refreshing. | |
| | |  | #2 |
Aleph Kids puts an AI language companion called Annie directly inside Minecraft.
Instead of telling children to close the game because it's time for an English lesson, Aleph more or less says:
Fine.
Let's put the English lesson inside the game.
Children enter a private Minecraft world where Annie talks with them, plays alongside them and gives them missions to complete using spoken English. The system is designed to make children speak repeatedly while playing rather than sitting through a traditional tutoring interface.
Aleph says its learning approach combines language acquisition ideas with Minecraft's existing exploration and building mechanics. The product is particularly pitched toward children who may be uncomfortable speaking in a face-to-face lesson but are happy chatting through an avatar while playing.
The company currently offers a seven-day free trial. Why we like it: this is one of those ideas that feels obvious after someone builds it. Most educational software asks children to leave the thing they actually want to do and enter an educational environment. Aleph does the reverse. It puts education inside the environment they already want to explore. And that's the more interesting Voice AI idea here. Voice doesn't have to be the product. Sometimes it can become an invisible interaction layer that makes another experience better. For education, gaming and kids' products generally, that feels like a pretty big playground. Pun slightly intended. | |
|
|---|
01 | Spotlight Rob Pickering, Founder & CEO @ Aplisay Founder conversation |
 |
| | How did you end up in voice?Long journey. Have been developing networking software and building Internet networks for decades. That lead on to Voice over IP as the next challenging horizon in the 2010s. Built a successful VoIP company which I sold in 2019 and could see Voice AI as the next frontier. Founded Aplisay in 2019 and have run it through the two major technology waves of trained intent recognition in 2019 to 2022 and LLMs from 2023 on. Just about to launch polite.ai which encapsulates everything we have learned. What's your struggle or moment of joy with voice?Biggest challenge: going too early. Tried to build an SME facing self service telephone assistant in 2020 using Lex/Dialogflow era technologies when both the tech and the market wasn't ready. Moment of Joy: rebuilding on GPT in early 2023, standing on a stage and demoing it to an incredulous voice industry audience. From that moment I knew, like the Internet and VoIP, this was a tech that was finally going to change the world! Where do you think voice is going?There are two fairly distinct application areas at the moment. The first is putting a voice interface on agents that work for us, just because we think closer to the speed we can talk than the speed we can type. This is largely owned by the model vendors and independent chat interface providers. The second is using voice agents as an outwards facing persona of our organisations. Right now, we see the second through a lens of human agent replacement. What is selling is AI bolted into telephony as a cost saver. These worlds will merge eventually, with voice (& video avatar) interfaces as the outward facing presentation of teams of agents implementing a business model. What that looks like commercially will probably be AI teams managing the whole communication processes, rather than AI agents bolted discretely onto the side of telephony systems. That is what I'm betting on anyway. | |
| | | The GLOBAL VOICE AI EVENT was a massive success |
On 20 August, Voice AI people came together in Melbourne, Bengaluru, Dubai, Istanbul, Paris, Barcelona, London, Toronto, and San Francisco. Nine cities, one day, one community showing up around the world. And honestly, it was a massive success. What started as a slightly ambitious idea turned into something pretty special, builders meeting builders, conversations spilling well beyond the schedule, and the very online world of Voice AI becoming wonderfully offline for a day. Nine cities down. Next time? Let's make the map a little more crowded. | |
| | Support & Share Enjoyed this newsletter? | |
| | Please share it with someone else interested in Voice AI. It helps us more than you might think, and it means we get to keep spending unreasonable amounts of time finding speech technology on the Internet.
If you're building something useful, weird, technically impressive or simply worth knowing about, send it our way. Same goes for Voice AI news, research, experiments and founders we should speak to.
We're always looking.
Have a lovely morning, day, evening or night wherever you might be.
The Voice AI Space team | |
|
