cell is like matrix , but each element could be from a different type, even a structure. If you see the example I gave the elements in column 1 is a string, while elements in column 2 could be a floating value for example (if fingerprint_of_one=2.3).
Let assume you calculate a short time energy of a speech signal (I suppose you know at least this one). You divide the recording on, let say 3 segments, then the vector with short time energies for "one" should look like:
fingerprint_of_one=[ 5.7 4.8 2.1];
doing the same for the word "two" colud result into some other valuse, for example:
fingerprint_of_two=[ 2.3 3.5 8.8];
then for an unknown word ([3.1 3.8 7.1]) you have to calculate the distance measure (euclidean in your case) to the two fingerpints and to check which one is the closest match.
Now, if you expand the example from my previouse post with todays vectors, type them in Matlab and press enter, they will be stored in Matlab's working memory under the variable "Vocabulary". They will stay there until you close Matlab. Therefore after the initialisation you may recognize any word by comparing its distance to the different fingerprints (Vocabulary{:,2})
Is it clear now? Hope you can get it, but actually you have to read a lot of books/papers to do such a system.