New blog, exploring DSP capabilities of the Arduino :
http://arduinovoicerecognition.blogspot.com/
I'm not a professional programmer , probably there are some bugs in sketch, as the say http://www.murphys-laws.com/murphy/murphy-computer.html. 
Would be nice to hear new ideas for improvement.
Current voice-print match probability above 40-50 % for password, and less than 20 % in any other case. The best results with two-words phrases, close to 1 sec in length, like "Thirty three" or similar, though content is very "speaker dependent" .