Research

Voices, made legible

I build models that recognize a speaker, describe them in language, and stay reliable when the audio is noisy, short, or adversarial. Below are the projects that define that program of work.