You can cook up your own routine without too much effort and it can be much more useful than the soundex routine. Here are some considerations in developing a soundex like key.
How many total records are you looking at that will need to be indexed? if you have 100 records, then the first letter of each word provides a manageable returned record set, but if you have 300,000 records, then you need something a bit more sophisticated. So, you need to choose a granularity... the average number of records that would be returned based solely upon the soundex code.
Are you planning on using this on the basis of how words sound or how they are written? If written, you will want to focus more on specific character matching between two encoded words (presence of the same consonants, for example), whereas if on how words sound, then you'll want to mash together similar sounding consonants like S and Z, or C and K. Also, you handle odd combos... like EAU as in BUDREAU (boodr O), EUAX, etc. IN other words, there are combinations of letters that have a common sound and these all get mashed to that common sound.
A common approach is to save the first letter, then substitute for common sounding strings, then mash out all vowels, then mash out all double consonants.
I wrote a soundex like routine that provides a maximum of a 6 character alpha key. It was designed for a dispatch operation where operators take the name over the phone and look up an address with it. The address database has about 105,000 unique names in it.
Now, this routine is not perfect, and it has never been rigorously analyzed for its statistical "togetherness", but it has been used quite sucessfully by the third largest sheriff's department in the United States.
If you'd like the code to modify or laugh at, email me.
