I am trying to recognise the alphabetical letters by saving some data of records of me and others saying each letter, then comparing a test audio not contained in the data to each letter using the intercorrelation. The correlation that has the maximum value must be the one corresponding to the letter recorded in the test. I did it using mathematica, some tests worked but others didn't. When I asked somebody, he told me to filter audio from noises. I don't know how to do it. Can somebody help me?