This thesis documents the development of a generic backend capable of finding the best match for audio fingerprints. It can be used with various fingerprint generators and audio sources subject to noise. The best matching fingerprint is determined by a scoring system, where the scores depend on the amounts of correctly positioned n-grams in the fingerprints. The developed algorithm uses n-grams and hash tables for fast lookup. Additionally, a similarity measure is developed to quickly create candidate sets. The candidate set contains a subset of the database fingerprints, that is estimated as a possible match. The search algorithm guarantees no false dismissals and parameters can be adjusted to alter the reliability of the results. Experimental performance studies shows that the solution is orders of magnitude faster than related work.