They have this feature implemented inside of Google Chrome (or maybe it is Webkit, I am unsure) and I take it that you're using Chrome to test this feature. As far as I know, it's an experimental API that they have added to the HTML5 specs. In the video in that blog post they say that you need to have the latest version of Chrome to use it : http://googleblog.blogspot.com/2011/06/knocking-down-barrier...

Yeah this is a Google Chrome specific feature. Chrome records your voice, uploads an mp3 to a Google server and returns the text. It is about the least efficient way to accomplish the task. Ridiculous really. Our operating systems (even Windows95) have had speech features forever, but it's implemented in a very clunky way. Instead there should be a standardized speech-to-text input, or JavaScript API where I can use my operating system's built-in speech features.

The built-in speech recognition in your OS isn't as good as Google's (and it may not even be there — think Linux or Chrome OS).

My quick experiments with the Chrome speech input says otherwise. It is both less accurate, and less useful than the built-in speech-to-text in MacOS. There exists speech systems for Windows and open source ones for Linux that are "good enough".

The point isn't really about accuracy, it's about usability. The way this is implemented in Chrome does not make it possible to use voice commands to do operations in a web browser. That's what we need. We don't just need a voice input for Google search.