Don't do accustic fingerprinting. Just use md5 checksums or some other hashing algorithm. It would work better, and not be exclusive to audio files.