One kind of voice recognition is called template matching. It starts when you read a prepared script to the software in your home computer. The software makes a map of the frequencies you use to speak each sound, and associates those maps with letters and letter combinations. Later, when you begin to dictate, the computer recognizes those sounds and brings up the appropriate characters on the screen. Sometimes software may have trouble identifying whether you meant "here" or hear."
Another -- the kind you'll encounter if you get an automated response system which asks you to speak your answers -- is called feature matching. In this case, the software looks for common patterns that are present in many types of voices. It analyzes the words you speak before it tries to match the words with the patterns it already knows. Because voices can vary so much in accents, and voices can be deep, high, loud or soft, this type of pattern matching is restricted to fewer words.
Image copyright (C) Signalogic, Inc., 1999-2004.