Back To Schedule

Workshop 2: Build Your Own Shazam

Audio Fingerprinting from First Principles

Audio fingerprinting is one of the most successful real-world applications of audio signal processing. It powers music recognition systems such as Shazam, enabling reliable identification of songs from short, noisy recordings in a few seconds. While many audio developers are familiar with audio fingerprinting in practice, the underlying techniques are often encountered only at a high level. In this hands-on workshop, participants will build a complete audio fingerprinting engine based on Avery Wang’s landmark 2003 Shazam paper. Starting with raw audio, we will progressively construct the identification pipeline: generating spectrograms, detecting robust spectral landmarks, creating constellation maps, producing landmark hashes, building a lookup database, and performing audio matching through offset clustering. Rather than treating audio fingerprinting as a black box, the workshop develops the underlying signal-processing principles and connects them to the system-level structure of a working audio fingerprinting pipeline. Using a subset of the GTZAN dataset and provided starter code, attendees will experiment with different parameters and observe how these choices influence matching behavior and retrieval accuracy