This document describes an approach for improving the speed of audio fingerprint searches in large audio databases. It proposes using a more compact representation of audio fingerprints that reduces the memory requirements, while still maintaining accuracy. The key steps are: 1) extracting fingerprints from audio clips by transforming them into spectrograms and filtering specific frequency bands, 2) further compressing the fingerprints using wavelet decomposition and selecting the most informative components, and 3) indexing the compressed fingerprints using min-hash to allow fast retrieval of similar fingerprints from the database. The approach aims to significantly reduce search time compared to existing audio fingerprinting systems, while achieving comparable accuracy.