Make wake word creation simple and accessible for everyone
Train custom voice activation models in under an hour. No ML expertise required. Works with Home Assistant, Rhasspy, OpenVoiceOS, and any custom project.
Supports 20+ languages — English, Spanish, French, German, and more
0
Models
0
Downloads
0
Wake Words
How it works
Three simple steps to create your custom wake word model
Step 1
Choose Your Wake Word
Type any phrase in any language. Short, catchy words work best — like "Hey Nova" or "OK Jarvis".
Step 2
Test Synthetic Voice
Preview your wake word with multiple AI voices before committing. Make sure it sounds exactly right.
Step 3
Set Training Parameters
Choose sample count, language mix, and training intensity. Or use defaults and go in one click.
Step 4
Send to GPUs for Training
We train your model on cloud GPUs and deliver a ready-to-use ONNX file in under an hour.
Recommended hardware
Devices the community runs wake words on — from plug-and-play voice assistants to $14 dev boards
Built with openWakeWord
Software that runs the wake words trained here — assistants, satellites and firmware from the community
Why OpenWakeWord?
Privacy First
Downloaded models run entirely on-device. No cloud connection needed for inference.
Lightweight & Fast
Sub-100KB models that run in real-time on Raspberry Pi, phones, and browsers.
100% Open Source
Built on OpenWakeWord. Apache 2.0 licensed. Customize and self-host.
Multilingual
Train wake words in 20+ languages. English, Spanish, French, German, Italian, Portuguese, and more.
Built for creators
From hobbyists to professionals, OpenWakeWord powers voice activation everywhere
🏠
Home Assistant & Voice Assistants
Use your custom wake word directly in Home Assistant via the openWakeWord integration, Rhasspy, OpenVoiceOS, or any open-source assistant. Drop in the ONNX model and you're ready to go — no cloud, no subscription.
🤖
Robotics
Voice-controlled robots and embedded systems
🎮
Gaming
Hands-free commands and game control
♿
Accessibility
Assistive technology and custom interfaces
Use it with your AI agent — no login needed
Connect our MCP server and your coding agent can check wake-word quality for free, train custom models (desktop and ESP32), and download the results — paying by card link or crypto, with no account. Ideal mid-project: “add a wake word to my assistant” becomes one prompt.
API & MCP documentation
All 13 MCP tools, the accountless REST API with copy-pasteable curl examples, tier pricing and payment rails — for building wake words into your own product or script.
Frequently asked questions
What is a wake word?
A wake word is a short phrase — like "Alexa" or "Hey Google" — that a small, always-listening model detects locally on your device. When it hears the phrase, it wakes your voice assistant to handle the command that follows. OpenWakeWord lets you train a completely custom one, like "Hey Jarvis" or your smart home's own name.
How do I create my own custom wake word?
Type your phrase, preview how it sounds with synthetic AI voices, and send it to our cloud GPUs. Training generates thousands of diverse voice samples in your language, teaches a model to recognize the phrase, and delivers ready-to-use ONNX and TFLite files — usually in under an hour. No ML expertise and no voice recordings needed.
How do I use my wake word with Home Assistant?
Download the .tflite version of your model, drop it into the /share/openwakeword folder, and restart the openWakeWord add-on. Your custom wake word then shows up in your Assist voice pipeline settings. Every model page includes step-by-step instructions.
Which languages are supported?
More than 20, including English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, and Chinese. You can even mix several languages in one training run so the same wake word is recognized across accents and pronunciations.
What is Piper, and can I create a voice with it?
Piper is the open-source text-to-speech engine used by Home Assistant. In our Voice Center you can record 10 short sentences and fine-tune a Piper voice on your own voice — training runs on cloud GPUs and stops automatically once the voice has converged. You get a standard voice.onnx that works anywhere Piper runs.
Can my assistant talk back in my own voice?
Yes. Put your trained voice.onnx and its .json config into the /share/piper folder, and the Home Assistant Piper add-on picks it up as a selectable text-to-speech voice for Assist.
What about ESP32 devices like the Home Assistant Voice PE?
Wake words that run directly on ESP32-S3 microcontrollers use microWakeWord — a different engine with tiny quantized models. Our sister site trains those: microwakeword.com
What hardware do I need?
A Raspberry Pi or any Linux box runs full openWakeWord models, while ESP32-S3 boards like the Home Assistant Voice PE or ReSpeaker Lite run tiny microWakeWord models on-device. See our recommendations: recommended hardware
Is my voice data private?
Training uses synthetic voices, so no recordings of you are needed for a wake word. Downloaded models run 100% on-device — no cloud, no subscription, no audio ever leaves your network. Voice Center recordings are used only to train your personal Piper voice, unless you explicitly donate them.