String matching technique(s) by converting to number?

Question

I have various length strings which are full of Base64 chars. Actualy they are audio recognition datas differs by song-to-song.

For easily comparing parts of those strings i divide them into 16-char sub-strings. (which is about 1 second of a song) But in some cases, i just can't compare these ones head to head.. i should be measuring them.

For example comparison with 'hellohellohelloo' and 'hallohellohelloo' should get a closer value then 'hellohellohelloo' and 'herehellohelloo' comparison.

Is there any algorithm or theorical

---

Edit: Sorry, i am new here :) And i couldn't make myself clear. Here are some comments that will make me clear and proposes an idea.

Comment 1:

Actually i know about Levenshtein distance, but the problem is every time i compare two strings i have to build comparison matrix and that makes searching process slow. If i can convert for example hello to 4444 and hallo to 4443 i can determine how close records i have for 'hello' by just indexing numerical values.

Comment 2:

Maybe i should determine a base constant-length string(s) and store distance values from them as the index values for string. It's just an idea?!

Answer 1

Levenshtein's distance will probably help you : http://en.wikipedia.org/wiki/Levenshtein_distance

It's usually pretty fast, and there are implementations in most modern languages too.

Answer 2

The Levenshtein distance might work for you. Also see the Wikipedia overview of edit distance.