AI/ML PROJECTS
Footstep Audio Generator
An end-to-end ML system that generates realistic, synchronized footstep sound effects from video - solving one of the most time-consuming and repetitive tasks in sound design.
The Solution
-
Detects footsteps and walking patterns directly from video
-
Generates context-aware footstep sounds (surface, footwear, character)
-
Synchronizes audio to visual contact points frame-accurately
-
Applies spatial processing (panning and attenuation based on screen position)
-
Outputs production-ready audio files (WAV or embedded in video)
​
Technical Approach
-
Built with Python, PyTorch, and Stable Audio, using LoRA fine-tuning on professional foley libraries. Designed custom evaluation methodology when standard audio metrics (FAD, CLAP) failed to correlate with professional quality standards.
SHOW REEL
MUSIC
For a full list of projects and episodes, check out Yejin’s IMDb page.
MUSICAL SOUND DESIGN / TECH ASSISTANT
ABOUT
Yejin Bang is a film composer and audio ML engineer specializing in generative models and perceptual evaluation for production audio systems.
​
The career began as a musical sound designer and technical assistant on Hollywood films including Child's Play, Godzilla, and Happy Death Day 2U. After transitioning to composing, work has spanned over 85 films, collaborating with directors from around the world. These projects have been recognized at 14 international film festivals.
​
Now developing machine learning systems for audio generation - addressing production challenges where technical implementation and perceptual quality evaluation both require domain expertise.
