EEG Earbud Biometric Authentication

An in-ear EEG device for personal identification, small enough to fit inside an earbud — led the team through a classical-ML classifier and its transition to deep learning.

Source repository is private

Led a team building a personal-identification system around EEG signal recorded from inside the ear canal — biometric authentication small enough to actually wear, rather than a lab-bench EEG cap.

CAD assembly animation of the earbud: the shell opens to reveal the sensor electronics, coin-cell power, and pin contacts packed into an earbud form factor.

CAD render of the finished earbud seated in a 3D model of a human ear, showing how the in-ear EEG sensor fits the ear canal. — The assembled bud seated in the ear canal.

The first working version used a classical machine-learning classifier built in MATLAB on hand-selected EEG features. The project’s harder problem turned out to be motion artifacts: an in-ear sensor moves with the wearer in a way a stationary EEG rig doesn’t, and those artifacts were swamping the signal the classifier depended on. That drove the project’s main evolution — moving from the hand-tuned classical classifier to deep neural networks that could learn to reject motion artifacts directly instead of relying on manual feature engineering to filter them out upstream.

No public repo for this one — it predates my current practice of publishing lab/personal code, so this description is drawn from my own project records rather than source I can link to.