jaxcore / bumblebee-hotword

A minimalist hotword / wake word for the web, based on Porcupine
Other
56 stars 6 forks source link

screenshot

Bumblebee Hotword

Bumblebee Hotword is a stripped down and repackaged version of the excellent Porcupine wake word (hotword) system. This requires no cloud services and is freely available to use under the Apache 2.0 license (GPLv3 compatible).

When Bumblebee Hotword is added to a web page, it listens to the microphone and emits an event when it hears any of the available hotwords.

For more information about the Bumblebee platform and its other features visit:

NodeJS Version

If you need hotword detection in NodeJS use bumblebee-hotword-node.

Examples

Install

Using npm:

npm install bumblebee-hotword

And include in your web project:

import Bumblebee from "bumblebee-hotword";

Porcupine Web Assembly Codebase

The porcupine webworker files must be manually served from your public html directory and the location must be specified using setWorkersPath() before starting bumblebee.

bumblebee.setWorkersPath('/bumblebee-workers');

Copy the entire directory here to your web server's public html or assets directory.

Quick Start

const Bumblebee = require('bumblebee-hotword');

let bumblebee = new Bumblebee();

// set path to worker files
bumblebee.setWorkersPath('/bumblebee-workers');

// add hotword
bumblebee.addHotword('bumblebee');

// set sensitivity from 0.0 to 1.0
bumblebee.setSensitivity(0.5);

bumblebee.on('hotword', function(hotword) {
    // YOUR CODE HERE
    console.log('hotword detected:', hotword);
});

bumblebee.start();

Note: browsers require user-interaction to start the microphone.

Hotwords

The hotwords available by default are:

Due to processing time it is recommended to only add the hotwords that need to be used. The hotword spoken can be retreived in the .on('hotword') event:

bumblebee.addHotword('bumblebee');
bumblebee.addHotword('grasshopper');
bumblebee.addHotword('hey_edison');
bumblebee.addHotword('porcupine');
bumblebee.addHotword('terminator');

bumblebee.on('hotword', function(hotword) {
    console.log('hotword detected:', hotword);
});

The Picovoice hotwords open source hotwords are freely usable under the Apache 2.0 license. Custom hotwords can be licensed from https://picovoice.ai.

Adding a Custom Hotword

Like the open source hotwords, the file must be in Porcupine's "_wasm.ppn" format.

You can use the tools/convert_ppn.sh shell script (uses the xdd command line program) to convert the PPN file to a JavaScript file:

sh tools/convert_ppn.sh my_custom_hotword_wasm.ppn

The resulting .js file can then be added to Bumblebee:

import my_custom_hotword from './my_custom_hotword'
bumblebee.addHotword('my_custom_hotword', my_custom_hotword, 0.5)

Sensitivity

Hotword detection sensitivity (0.0 to 1.0) is configurable only before the first call to bumblebee.start()

bumblebee.setSensitivity(0.8);

Disable Bumblebee

Use the stop() method to disable the microphone and all processing:

bumblebee.stop();

Mute

bumblebee.setMuted(true); // mutes microphone volume

Set Microphone Volume

bumblebee.setMicVolume(0.5); // sets microphone volume to 50%
bumblebee.setMicVolume(2); // sets microphone volume to 200%

Spectrum Analyzer

Bumblebee Hotword instantiates an audio analyser which can be used to draw an oscilloscope or other types of processing:

bumblebee.on('analyser', function(analyser) {
    // analyser is an instance of audioContext.createAnalyser()
});

Audio Data

Bumblebee Hotword emits a stream of "data" events which can be used to receive the microphone audio data after downsampling to 16bit/16khz PCM. This is the format also used by other speech processing systems such as DeepSpeech.

bumblebee.on('data', function(data) {
    console.log('data', data);
});

Run Examples Locally

Clone this repo, then...

For the basic example:

cd examples/basic-example
yarn install
yarn start

For the full example:

cd examples/full-example
yarn install
yarn start

License

This repository is licensed under Apache 2.0. See Porcupine for more details.

Change Log

0.2.1:

0.2.0:

0.1.0: