When people hear that my Master's thesis in Artificial Intelligence is an iPhone app that plays noise, they usually wait for the punchline. There isn't one. There's a person: my husband has tinnitus, and for him silence isn't silent. I watched him try everything — medicine, acupuncture, every masking playlist the internet had to offer — and I couldn't make the ringing stop. So I picked the one tool I had and enrolled in an AI Master's to ask a narrower, more honest question: can a machine learn which sound gives one specific person their quiet back?
Why this is a machine learning problem
Tinnitus is deeply personal. The masker that saves one person's night does nothing for the next, and even for the same person, what helps depends on pitch, loudness, and the texture of the noise. That's a search problem over a parameter space, with a human in the loop providing noisy, expensive feedback — exactly the setting Bayesian optimization was made for.
MyTinnitusMask starts the way an audiology clinic would: a guided psychoacoustic session finds your tinnitus pitch, its loudness, and the minimum masking level — adapted for AirPods instead of a sound booth. Then each session plays two candidate maskers, A/B style, and you rate the relief each one gives. Comparing two sounds in the same session filters out the noise of good days and bad days. Behind the ratings, a Bayesian bandit (Thompson sampling) updates its beliefs after every comparison and converges on the masker parameters that work best for you. All of it runs on-device — there is no server, no account, no analytics.
Why it's free
Because it was built out of love rather than for profit, it stays free forever — no subscription, no ads. The research side follows the same principle: contributing a session is strictly opt-in, truly anonymous, and the dataset and methods will be published openly alongside the thesis.
Building for one person
The best design constraint I've ever had is that the first user sleeps next to me. Every shortcut a product team might argue about — "do we really need background audio with the screen locked?" — answers itself when you know exactly who is lying awake at 2 a.m. Building for one person you love turns out to be a surprisingly good way to build for the one in seven adults who live with the same sound.
The app is currently in beta on TestFlight, with an ethics committee submission and a small clinical pilot ahead. You can read more on the project page — and if your silence rings, or you love someone whose silence does: you are not alone.