on webkit-browsers you can use the get user media api with webkitGetUserMedia - like demonstrated on html5rocks.
if you want to use your voice to create javascript events (e.g. to control objects on the screen) you would have to analyze the incoming sound (e.g. high frequency for event1 - low frequency for event2, speech analysis is much more complicated, see below)
alternatively, there is chrome's 'x-webkit-speech' speech recognition, (see an example here), which will analyze speech on google's servers and is therefore probably too slow for real-time control.
I don't know of any real-time speech analysis in the browser, but would be happy to find a (even very basic) possibility.
edit: adds some code (adapted from here)
var mike = document.getElementById('mike');
mike.onwebkitspeechchange = function(e) {
console.log(e); // SpeechInputEvent
console.log(e.results[0].utterance);
};