Hark is a tiny browser/commonJS module that listens to an audio stream, and emits events indicating whether the user is speaking or not.
With browserify:
npm install hark
Without browserify download and use:
npm install hark
If you aren't using browserify, you'll want hark.bundle.js.
var hark = require('../hark.js')
var getUserMedia = require('getusermedia')
getUserMedia(function(err, stream) {
if (err) throw err
var options = {};
var speechEvents = hark(stream, options);
speechEvents.on('speaking', function() {
console.log('speaking');
});
speechEvents.on('stopped_speaking', function() {
console.log('stopped_speaking');
});
});Hark uses the webaudio API to FFT (get the power of) the audio in the audio stream. If the power is above a threshold, it's determined to be speech.
var speech = hark(stream, options);
speech.on('speaking', function() {
console.log('Speaking!');
});- Pass hark either a webrtc stream which has audio enabled, or an audio element, and an optional options hash (see below for options).
- hark returns an event emitter with the following events:
speakingemitted when the stream appears to be speakingstopped_speakingemitted when the audio doesn't seem to be speakingvolume_changeemitted on every poll event by the event emitter with the current volume (in decibels) and the current threshold for speech
- The hark object also has the following methods to update the config of hark. Both of these options can be passed in on instantiation, but you may wish to alter them either for debug or fine tuning as your app runs.
setInterval(interval_in_ms)changesetThreshold(threshold_in_db)change the minimum volume at which the audio will emit aspeakingevent
- hark can be stopped by calling this method
stop()will stop the polling and events will not be emitted.
interval(optional, default 100ms) how frequently the analyser polls the audio stream to check if speaking has started or stopped. This will also be the frequency of thevolume_changeevents.threshold(optional, default -50db) the volume at whichspeaking/stopped\_speakingevents will be firedplay(optional, default true for audio tags, false for webrtc streams) whether the audio stream should also be piped to the speakers, or just swallowed by the analyser. Typically for audio tags you would want to hear them, but for microphone based webrtc streams you may not to avoid feedback.audioContext(optional, default is to create a single context) If you have already created anAudioContext, you can pass it to hark to use it instead of an internally generated one.
Fine tuning the volume threshold is the main configuration setting for how this module will behave. The level of -50db have been chosen based on some basic experimentation on mysetup, but you may wish to change them (and should if it improves your app).
What is dB? Decibels are how sound is measured. The loudest sounds on your system will be at 0dB, and silence in webaudio is -100dB. Speech seems to be above -50dB depending on the volume and type of source. If speaking events are being fired too frequently, you would make this number higher (i.e. towards 0). If they are not firing frequently enough (you are speaking loudly but no events are firing), make the number closer to -100dB).
Clone and open example/index.html or view it online
- Chrome 27+, remote streams require Chrome 49+
- Firefox
- Microsoft Edge, support for remote streams is under consideration
MIT
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent