How to Add Custom Wake Words to any Web App

🚀 Best-in-class Voice AI!

Build compliant and low-latency AI apps running within web browsers without sending user data to 3rd party servers.

Learn how to add wake words, like Alexa or Hey Siri, to any web app. This tutorial takes 15 minutes or less from the start to a working demo. We learn how to train custom wake word models, like Hey Jarvis, that fit your product, not Big Tech's brand. In this article, we use Picovoice Porcupine Wake Word Engine Web SDK.

Wake Word Detection is also known as Keyword Spotting, Hotword Detection, Always-Listening Voice Commands, Trigger Word Detection, and Voice Activation.

Setup the Project

Create a new folder and initialize an NPM project:

npm init

Install the dependencies:

npm install @picovoice/web-voice-processor @picovoice/porcupine-web

Install http-server as a development dependency:

npm install http-server --save-dev

Download the Porcupine model (i.e. Deep Neural Network). From the project folder, run the following to turn the binary model into a base64 string. Remember that you need to replace ${DOWNLOADED_MODEL_PATH} with the path to the model you downloaded (e.g. ~/Downloads/porcupine_params.pv on my Ubuntu machine).

npx pvbase64 -i ${DOWNLOADED_MODEL_PATH} -o porcupine_params.js

Create a boilerplate HTML file called index.html with the content below:

<!DOCTYPE html>
<html lang="en">
<head>
  <script src=
    "node_modules/@picovoice/web-voice-processor/dist/iife/index.js">
  </script>
  <script src=
    "node_modules/@picovoice/porcupine-web/dist/iife/index.js">
  </script>
  <script src="porcupine_params.js"></script>
</head>
<body>
<h1>Porcupine Web Demo</h1>
</body>
</html>

Run the local server to load the page:

yarn run http-server -a localhost -p 5000

You can see the page at http://localhost:5000.

Train Wake Word Models

Sign up for Picovoice Console.
Go to the Porcupine Page.
Select English as the language for your model.
Type in Hey Jarvis as the phrase you want to build the model for.

Optionally, you can try it within the browser
Once you are happy, click on the train button.
Select Web (WASM) as the platform.

Click on Download. You should have a .zip file in your download folder now.
Unzip it. Inside the folder, you see a file with the suffix .ppn. That's our model. Transform it into a base64 string. Remember that you need to replace ${DOWNLOADED_PPN_PATH} with the path to downloaded file (e.g. ~/Downloads/Hey-Jarvis_en_wasm_v2_1_0/Hey-Jarvis_en_wasm_v2_1_0.ppn on my Ubuntu machine)

npx pvbase64 \
-i ${DOWNLOADED_PPN_PATH} \
-o hey_jarvis.js \
-n HEY_JARVIS_KEYWORD_MODEL

Wire it up

Go to Picovoice Console's dashboard. Copy your AccessKey.

Add the base64 keyword model as a script:

<script src="hey_jarvis.js"></script>

Add div elements for debugging and showing the result:

<div id="status"></div>
<div id="result"></div>

Initialize Porcupine and start listening to the microphone. Remember to replace ${ACCESS_KEY} with your AccessKey copied from the Picovoice Console.

<script type="application/javascript">
function porcupineErrorCallback(error) {
  console.log(error);
  document.getElementById("status").innerHTML = error;
}

function porcupineKeywordCallback(detection) {
  const time = new Date();
  const message = `Detection at ${time.toLocaleTimeString()}`
  console.log(message);
  document.getElementById("result").innerHTML = message;
}

async function startPorcupine() {
  try {
    let porcupine = await PorcupineWeb.PorcupineWorker.create(
      ${ACCESS_KEY},
      [{label: "Hey Jarvis", base64: HEY_JARVIS_KEYWORD_MODEL}],
      porcupineKeywordCallback,
      {base64: modelParams},
    );
    await window.WebVoiceProcessor.
      WebVoiceProcessor.subscribe(porcupine);
  } catch (err) {
    porcupineErrorCallback(err);
  }
}
</script>

Additional Languages

Porcupine supports many more languages aside from English. To use models in other languages, refer to the quick start.

Source Code

The source code for a fully-working demo with Porcupine is available on its GitHub repository.