A comparative analysis of audio fingerprinting techniques as applied to modified content-based audio identification

• Main Author: Lachica, Daniel Philippe F.
• Format: text
• Language: English
• Published: Animo Repository 2022
• Subjects: Information storage and retrieval systems—Fingerprints; Databases and Information Systems; Theory and Algorithms
• Online Access: https://animorepository.dlsu.edu.ph/etdm_comsci/14; https://animorepository.dlsu.edu.ph/cgi/viewcontent.cgi?article=1017&context=etdm_comsci
• Institution: De La Salle University

Audio fingerprinting techniques are commonly used to programmatically generate unique, compact digital signatures for songs. Given a fingerprint database of substantial size, these algorithms are capable of identifying a plethora of songs across a wide range of genres and languages based on a few short, contiguous seconds of auditory input. Existing studies point toward the use of audio fingerprinting algorithms for content-based audio identification. However, little is known about the relative performance of these algorithms when the audio file input has been intentionally tampered with, as in the case of audio modification for purposes of either unjust duplication or copyright infringement. On that premise, the goal of this study is to provide a comparative analysis of the performance be- tween two audio fingerprinting techniques, namely the Shazam and Quad-Based (Qfp) algorithms, as applied to the task of modified content-based audio identification. Performance indicators show that the Qfp algorithm outperforms the Shazam algorithm across all standard and modified audio identification tests on a collection of distorted audio samples built from high-quality music files. More studies are needed before drawing conclusions about the most ideal content-based fingerprinting approach to modified audio identification, and perhaps integrity verification tasks with mixed-signal audio queries.