When you are at a cocktail party the background noise is significant, actually most of the time it is so loud that it is difficult to hear individuals talking. And yet, I am sure you had the experience of actually talking with one of the guests and being able to hear him/her.
It is like, with some magic, your ears/brain managed to filter the background noise and focus on the only sounds that matter to you. And actually this is what happens in your brain. Scientists call this capability of focussing on a specific meaningful sound masking all the rest "cocktail party effect".
And this is what scientists at the Duke University have managed to implement in a sensor enabling computers to focus on a sound stream cutting off all the others.
They have created a sensor in the shape of a pie with a honey-comb base split into dozens of slices. Each comb, in the honey-comb, has a different depth and this affects the sound waves. Each slice has a different pattern of honey-comb, hence sound entering a slice will get a very specific distortion, a specific signature, that a computer can single out to distinguish it from sounds coming from other directions. This makes for a very effective sound-homing sensor.
The prototype developed is pretty big, about six inches, too big to be incorporated into a smart phone. However, the team feels they will be able to scale it down, still keeping its characteristics, to the point that it can be inserted into a smartphone. This will considerably improve the usability of voice detection since the software will be able to distinguish between your voice and other voices.
The invite from Apple to their September event has the headline "Hey Siri, gives us a hint" seems to suggest a greater role in voice based computer interaction, and I bet this will bloom over the next few years. May by the keyboard I am typing on right now will be a thing of the past in the next decade...