Cataloguing the Chaos
An Embedded-Based Auto-Tagging System for SFX Libraries
Managing third-party SFX libraries is one of post-production's most time-consuming tasks. Sound designers and audio programmers routinely struggle with inconsistent vendor folder structures, missing metadata, and arbitrary naming conventions.
Existing systems suffer from two main gaps: they rarely match up how different vendors name things into a single taxonomy, and their all-or-nothing automation either collect hidden metadata debt or manual review for every track by the user.
The project handles both cases. If a file already has some metadata, even messy or inconsistent, it gets cleaned up and mapped onto the UCS standard. If it has none at all, an audio classifier steps in instead: trained to recognise common categories, with a zero-shot fallback for anything it hasn't seen before. Every predicted tag comes with a confidence score: high-confidence tags get applied automatically, and anything uncertain gets flagged for the sound designer to check.
Denise Azucena
Software Developer - Machine Learning
Independent
Denise is an audio engineer with over 5 years of experience in adverts, short films, and audiobooks. She holds an MSc in Sound and Music Computing from Queen Mary University of London, where she focused on AI and data science. She now works as a freelance software developer focused on machine learning, with a strong interest in audio programming and sound design.
Combining her deep domain knowledge of sound with machine learning and audio information retrieval, she draws on her post-production experience to build software tools that address real pain points in the industry.