ARM Cortex-M4

It is incredibly precise engineering.

The implant they inserted contained an Arm Cortex M4 microcontroller.

Which is a tiny but very capable little processor.

Exactly.

When is nearest centroid sufficient? The nearest centroid classifier in the L two MFCC space is viable in scenarios with a small number of speakers, two to four, relatively controlled acoustic conditions and latency requirements under fifty milliseconds.

Execution in the browser or on Arm Cortex M4 plus microcontrollers.

On the VoxCeleb one dataset, seven thousand plus speakers nearest centroid with MFCC yields an EER of approximately fifteen to twenty percent, compared to one to three percent for DNN models.

The gap narrows significantly when the number of speakers is small and training data is abundant: one, three.