I miss my father's voice, so I'm building it back

I miss my father’s voice.

Not the idea of it. The actual sound. The way he says my name wrong on purpose.

Like most people, I always assumed I would record him properly someday. What I actually have is what everyone has: noisy home recordings. A fan in the background, three people talking at once, a phone microphone from years ago.

This weekend I ran those recordings through modern voice AI, and they broke. Not because the models are weak. The models are astonishing now, and they run entirely on a Mac. They broke because the recordings were damaged. Voice preservation turns out to be an archaeology problem, and archaeology fails when the artifacts are cracked.

So I started building the missing tool. A Mac app that treats a voice like a keepsake: it takes any old recording, splits it into usable moments, lets you trim and repair them on device, removes the background noise, and builds a voice profile that can say new things. In that voice. In that tone.

If someone you loved is already gone, I want to say this part gently. The recordings you have are not too broken to matter. A voicemail, a shaky birthday video, thirty seconds of them talking over everyone at dinner. Bring them as they are. The repair room in this app exists precisely for the recordings nobody planned to make. And if you never recorded them at all, that is not a failure. Nobody teaches us to archive the people we love. We learn it the hard way, one silence at a time.

I also want to be honest with the people this technology unsettles. You are not wrong to feel that. A voice is the most intimate thing a person leaves behind, and treating it casually would be a kind of theft. That is why the ground rules are built in, not bolted on. Every voice starts with consent recorded in the person’s own words. Every generated clip carries provenance and identifies itself as synthetic. Revoking consent disables the voice instantly. Nothing ever leaves the machine. And nothing plays until you press play. Grief has its own schedule, and software should respect it.

Here is the part that still feels like science fiction to me. I did not write most of this app. The AI agents I have been building for Storyline did, while I directed. The pipeline, the audio repair room, the consent system, the app itself, built in days, not months. Building the agent infrastructure first is paying for itself in ways I did not predict.

It will ship as a standalone app. Local only. The voice is a folder you can open in Finder. No cloud, no accounts, no subscription, and it works in airplane mode.

I keep thinking about who this is really for. Children of aging parents. Families split across oceans who hear each other mostly through phone speakers. People facing an illness that will take their voice while they are still here, who deserve to bank it on their own terms. And anyone who just wants their grandmother’s laugh to outlive the phone it lives on.

So, one request, whatever you think of AI. If your parents are alive, record ten clean seconds of them today. A quiet room, one story, their real laugh. Storage is free. The window is not. And if they are already gone, gather what you have this week. It is more than you think.

Still early. The name is not final and my father’s profile is not finished. But it is closer than it was on Friday, and this time the recordings will be good enough.