A Pocket Sphinx binding for Node.JS, focusing on continuous recognition.
This repo is experimental and out-of-date. I strongly recommend checking out the much better supported cmupsphinx fork.
pocketsphinx requires CMU's Pocket Sphinx libraries. You can build and install it yourself, but I highly recommend just getting it from a package manager. For Mac OSX I highly recommend homebrew. It makes this a snap:
brew install cmu-pocketsphinx
A PocketSphinx
object is a writable stream that accepts mono 16 kHz 16-bit PCM data. When an utterance is detected, e.g. a phrase, word, or statement, the object emits an utterance
event with the details of the detected utterance.
var PocketSphinx = require('pocketsphinx');
var ps = new PocketSphinx();
ps.on('utterance', function(hyp, utt, score) {
console.log( 'Guessed phrase: ' + hyp);
console.log( 'Confidence score: ' + score);
console.log( 'Unique utterance id: ' + utt);
});
ps.write(myMicrophoneData);
Data can be retrieved from files, but the demo shows how to use getUserMedia and Socket.io to stream data from a web frontend. This is also a good demo of how to use getUserMedia for cool stuff, considering the documentation for it is a little sparse.
This is still in active development. It has only been tested on Mac OSX 10.9.2 with Pocket Sphinx installed using homebrew.