From Human Voice to Drone Detection: One Mathematical Fingerprint.

From Human Voice to Drone Detection: One Mathematical Fingerprint. What do a doctor–patient conversation and a drone flying over critical infrastructure have in common? Surprisingly, they can be analyzed using the same mathematical principles. This year presentation explores real-time speaker diarization directly in the browser using Web Audio API and a lightweight machine learning pipeline: FFT → Mel Filterbank → MFCC → L2 Normalization → Nearest-Centroid Classification. We will discuss practical implementation challenges, key design decisions, and why lightweight classifiers can sometimes outperform more complex approaches in constrained scenarios. The most interesting part? With a few parameter adjustments, the same pipeline can be adapted to identify drones based on the acoustic signatures of their motors and propellers. One pipeline, two very different applications, and a shared mathematical foundation. An exploration of Digital Signal Processing, Machine Learning, Web Technologies, and dual-use AI applications.