Do we really want to talk to a web page? Before you answer think about how many users talk to their mobile. Now there is a standardised API for it.

Hot on the heels or should that be tail of the Firefox 4 release Google Chrome has added some interesting extras to its beta channel.

Google not only seems to be the first to support advanced browser features but it seems to be set on pushing all things HTML ever further into what we used to regard as desktop territory. Now we have speech recognition added to HTML5. After working with the Incubator group a draft standard has been prepared for the Speech Input API.  This basically lets a user talk to a web page rather than type.

Speech recognition has been available on most desktops in the form of the Microsoft Speech SDK and SAPI - but you don't see it much in use. There could be a reason for the under use of this facility in that speech recognition has never achieved an accuracy that makes it worth using unless you go to a great deal of trouble to train it and work in a quiet room.

However speech recognition may be staging a comeback but not on the desktop - on the mobile phone. Many mobile phone users are very happy using speech input to select phone numbers to dial and to control their phone in hands off situations - usually in a car. So perhaps speech recognition on the web isn't doomed to failure but it will need a good idea, careful design and implementation.

The web SDK is remarkably easy to use - all you need is a new tag:

http://www.google.com/search">
<input type="search" name="q" 
  speech required 
   onspeechchange="startSearch">
</form>

Now the user simply clicks on the input element starts talking and the recognised text is submitted as the content of the input element - as if it had been typed.

There is a demo that you can try out: http://slides.html5rocks.com/#speech-input

but it only works if you have the correct beta of Chrome. Recognition is performed by a speech server but details of this are vague at the moment. Indeed the entire specification is at a very early stage.

You can also provide a simple grammar that the input has to correspond to to both limit the range of possible inputs and to improve the accuracy. For example, if your application has only a limited range of commands that it can respond to then a grammar limiting the recognition results to the commands is a good idea. 

As well as the speech API the latest beta of Chrome also supports the GPU accelerated 3D CSS which allows you to add 3D effects by applying a style. It is clear that the browser war is hotting up despite or perhaps because of the recent releases of IE9, Chrome  10 and Firefox 4.

More Information

Chrome beta channel

HTML Speech API spec

Incubator group