A Speech Recognition Solution to an Ancient Cryptography Problem

The common substitution cipher that shifts every letter in the alphabet n letters to the left or right is known to every cereal box reader. Perhaps less known is its extension to poly-alphabetic substitution ciphers invented in 1553 where consecutive letters are shifted by different amounts. This can be according to a periodic key or a one-time pad, introduced in 1917. The randomly generated non-repeating key, i.e. one-time pad or running key cipher, is not possible to decipher. However, if the non-repeating key is text, a common encryption mistake, then the message can occasionally be deciphered. In this paper we show how to decipher using the Viterbi algorithm. The case of repeating keys, both English text or random, will be treated as well. We measure the error rate of the decryption on a test set where the messages and keys are known. We further resolve a cipher text for which the original message and key was not available to us.