Last night I went and saw the brilliant ventriloquist Nina Conti. The show was hysterically funny. One of the recurring jokes is the sending up of the art of ventriloquism, about how the audience knows that Nina is voicing both parts and yet we go along with it. This YouTube clip gives a great example at the top of the set:
What is amazing is that the lack of the microphone on Monkey doesn’t screw up the illusion, even when Nina draws attenntion to it. You feel like the sound is coming from Monkey’s moving mouth, even though the speech is actually coming out of Nina’s mouth (and on YouTube is actually emerging from your computer’s loudspeakers or through your headphones).
This illusion happens because there is an area of the brain that processes both sound and vision, allowing what you see to influence where sounds appear to be coming from. As neurobiologist Jennifer Groh commented in 2007: 
What we see also plays a vital role teaching us where sounds are located.  Scientists investigating this Ventriloquist Effect have found you can use this to retrain people to localise things incorrectly for sometime. Typically scientists flash a light onto a screen and tell their subjects this is where the sound is coming from. In reality, the loudspeakers hidden behind the screen are producing sound from a slightly different position. 20-30 minutes of training with the light leading the subjects to think the source is misplaced (say a few degrees to the right), is enough to temporarily recalibrate the sound localization system in the brain.