I recently built a project that allows users to send voice commands to their Yun using an HTML5 website. You can find a full tutorial and demo video for it here.
The project allows users to browse to a web page hosted by the Yun on any device, then speak in commands to change the color of an RGB led connected to the Yun.
The web page includes a simple text input field that is enabled with HTML5's Speech Recognition API, which provides Javascript functionality for accessing the browser's audio stream, automatically detecting speech, and returning translated speech input as a string.
As the user speaks out commands, the API tranlsates them into a string. This string is then passed to the Arduino sketch using the Yun's Mailbox functionality. The string is appended to the url http://arduino.local/mailbox/ and a request is made to it using jQuery.
A sketch running on the Yun is listening for Mailbox messages. As they come in the messages are parsed out, and in this case, it looks for names of a few colors in the command string. When it recognizes these colors the sketch changes the RGB to that color.
This method of voice recognition could be useful for many applications because it allow for any words to be translated accurately and sent to the Yun - its just up to the sketch to parse out the string to determine what action to take. The translation is often very accurate as well.