Use case · Mobile & consumer products

Give your product its own "Hey YourBrand" moment

A branded wake word turns your product into something users talk to — "Hey Sonos", "OK Google", "Hey Siri" all started here. With VoxRT it runs entirely on the device: no cloud listening on your users, no per-user fees on your margins, and battery cost too small to notice.

Hey Vox 0.97
Why teams do this

What a branded wake word buys you

A brand moment users repeat daily

Every activation is your brand name, said out loud, by choice. No other part of your product gets spoken hundreds of times a month — and once users learn the habit, it travels with them across your lineup.

Hands-free engagement

Cooking, driving, training, working — the moments users most need your product are often the moments they can't touch it. A wake word keeps it reachable when hands and eyes are busy.

A privacy story you can defend

"Always listening" is only scary when audio leaves the device. On-device detection means the microphone stream never goes anywhere — a one-sentence answer for store review, security review, and your users.

Live demo · runs in your browser

Start demo and say "Hey Assistant"

Your audio never leaves this page. The model highlights "Hey Assistant" while ignoring all other speech — including similar-sounding words and phrases like "Hey sister" or "Assist me".

Open the full demo →

"Hey Assistant"

Start speaking…
0.90
Where it shows up

The same feature, across very different products

Anywhere hands or eyes are busy — or a product simply deserves to answer to its own name.

Fitness & wearables

"Hey Coach, start interval two"

Mid-workout, hands sweaty or gloved — the training app responds without a single tap, even on a watch with no keyboard.

Automotive & mobility

"Hey Drive, save that clip"

Dashcams and driving companions where eyes stay on the road — and where cellular coverage can't be assumed.

Cooking & recipe apps

"Hey Chef, next step"

Flour-covered fingers never touch the screen. The classic wake-word moment — users learn it in one meal.

Smart home & audio hardware

"Hey Aura, turn it down"

Speakers, soundbars, and appliances that answer to their own brand name — without routing living-room audio through someone else's cloud.

Kids' toys & education

A plush toy that knows its name

The toy responds when a child calls it — and because nothing is recorded or uploaded, it's a privacy story parents actually accept.

Healthcare & assisted living

"Hey Nurse, I need help"

Hands-free calls for assistance from a bedside tablet or wearable. No audio leaves the device, which keeps compliance reviews short.

Field service & industrial

"Hey Rig, open the checklist"

Gloves, ladders, and machine noise rule out touchscreens. Rugged Android handhelds and headsets trigger workflows by voice.

Kiosks & hospitality

Walk up and say the name

Ordering kiosks and hotel-room devices that wake when greeted — no grimy touchscreen required, no cloud account per device.

Accessibility

The product's front door, by voice

For users with limited mobility or vision, a wake word makes the whole product reachable without locating a button first.

How it works

From phrase to shipping feature

Pick the phrase

Usually "Hey" + your brand, but single words and multi-word activations work too. Three-plus syllables perform best — longer phrases are harder to confuse with everyday speech.

We train and tune the model

VoxRT trains a compact detector for your exact phrase and tunes it against "hard negatives" — words that sound like your phrase but aren't — for your languages and target devices. You receive a model file about the size of a small photo (~100 KB).

Your team drops in the SDK

A native Swift Package on iOS, a Gradle dependency on Android, or a plain C library on Linux devices. The always-on listener is gated by voice activity detection, so it sips power while idle. Integration is a day of work, not a sprint — and teams usually build against the free "Hey Assistant" reference model while the custom one is in training.

Tune the trigger to your product

One threshold setting trades "never misses" against "never false-triggers". A meditation product biases toward never interrupting; a car-mode product biases toward always answering. You choose, per product.

For technical evaluators

The numbers behind the promise

Everything below is measured on real hardware and published in the GitHub repos — the same numbers, methodology and thresholds. See the benchmark methodology for how they're produced.

~1.5%
of one CPU core during continuous listening (iPhone 13 Pro Max; ~2% on a 2020 budget Android)
0.993
precision at the default threshold — 0.5% false-positive rate on a hard negative test set
~100 KB
wake-word model on disk; ~600 KB net APK impact on Android, 2–3 MB on iOS
0
bytes of microphone audio leaving the device — detection is fully local

Pipeline. Microphone → on-device VAD (gates everything; ~1 ms per 32 ms frame) → wake-word detector (runs only on speech) → threshold-crossing event to your code. Both models ride the same VoxRT runtime — one dependency, one binary.

Platforms. iOS 16+ (Swift Package Manager) and Android 8.0+/API 26 (JitPack/Gradle), plus Linux aarch64 for device-class hardware — validated on Raspberry Pi 3/4/5 and Pi Zero 2 W, NVIDIA Jetson Nano and Orin Nano, AWS Graviton and similar Cortex-A53/A55 boards — shipped as a single ~480 KB shared library with Python, Node.js, Go, C/C++ and Rust wrappers. Quickstarts with a live-microphone loop are in the docs: iOS · Android · Linux.

Licensing. The SDK and reference model are free for commercial use with no per-user or per-device fees; the custom phrase is a one-time tuning engagement. Details on the licensing page.

FAQ

Branded wake words, answered

How long does it take to add a wake word to a product?

The SDK integration is hours, not weeks — a Swift Package, Gradle dependency, or C library plus a short audio-capture loop, with quickstarts in the docs. A custom phrase adds the model-tuning engagement on our side; the free reference model ("Hey Assistant") lets your team build and test the full integration before the custom model arrives.

Will an always-listening wake word drain the battery?

No — this is what on-device wake words are designed for. VoxRT's detector uses about 1.5% of one CPU core on an iPhone 13 Pro Max and ~2% on a 2020 budget Android, and it's gated by voice activity detection so it idles when nobody is speaking.

Does it send audio to the cloud?

Not with VoxRT: detection runs entirely on the device, microphone audio never leaves it, and the feature works with no network connection. Nothing is uploaded unless your product decides to do so after the wake word fires.

What does a custom branded wake word cost?

The SDK, runtime, and reference model are free for commercial use with no per-user or per-device fees. The custom phrase itself — trained and tuned for your brand name, languages, and target devices — is a one-time paid engagement, not a recurring per-user license. See licensing.

Ship "Hey YourBrand"

Tell us your phrase and which devices it has to run on — build against the free reference model today.

Get started Try the browser demo