I didn't set out to build a dictation app. I was just a guy who discovered how much faster it was to talk than to type, got completely hooked on it, and then watched the tool I depended on fall apart.
Getting hooked on Willow Voice
It started with Willow Voice. I'd been skeptical of voice dictation — Apple's built-in option was underwhelming, and I assumed everything else was the same. But a friend convinced me to try Willow, and within a week I was using it for everything. Emails, Slack messages, notes, even long-form writing. Speaking at 150 words per minute instead of typing at 80 was a game-changer.
I was paying $12/month for it and it felt worth every penny. The transcription was fast, the accuracy was good enough, and the push-to-talk workflow felt natural. I was a convert.
Then things started breaking
A few months in, Willow started having growing pains. Latency crept up — what used to take under a second started taking two, three, sometimes five seconds. For a tool you use dozens of times a day, even small delays are maddening. You'd finish speaking and just... wait. Staring at a spinner while your thought evaporated.
The root cause was obvious: cloud processing. My voice was being sent to a remote server, transcribed there, and sent back. When the service was under load or having infrastructure issues, everything slowed down. Some days it was fine. Other days it was unusable. I started typing instead of talking on the bad days, which defeated the whole point.
I also started thinking more about privacy. Every word I spoke was leaving my machine. Work emails, personal messages, half-formed ideas — all of it hitting someone else's server. I'd been so focused on the speed that I hadn't really sat with that.
Trying the alternatives
So I started looking around. I tried SuperWhisper, which is a well-built app with good transcription quality. But it's another subscription ($10/month), and the setup asks you to choose between Whisper model sizes — small, medium, large, English-only vs. multilingual. If you know what that means, fine. If you're a normal person, it's confusing. It also has a cloud mode, which I wanted to avoid entirely.
I tried MacWhisper, which runs locally and has fair one-time pricing. But it's really a transcription tool — import an audio file, get a transcript. For live "hold a key and talk" dictation across any app, it's not the right fit.
I looked at Wispr Flow, which does interesting AI-powered text rewriting, but it's fully cloud-based. That was a non-starter after my Willow experience.
Nothing did exactly what I wanted: fast, local, dead-simple, works everywhere, no subscription, no configuration, no cloud.
Building the thing I wanted to use
I'm a developer, so I started building. The core idea was simple: run a speech recognition model directly on the Mac using Apple's CoreML framework. No server. No network request. Audio goes from the microphone into the model and out comes text — all on-device, all in under a second.
I built it for myself first. Push-to-talk with a configurable hotkey. Text gets pasted wherever your cursor is — any app, any text field. Smart Cleanup to strip out the "ums" and "uhs" that make raw transcription look sloppy. The transcription model downloads once on first launch and then everything works offline.
After using it daily for a while, I realized other people probably had the same frustrations I did. Paying monthly for something that should run locally. Dealing with latency from cloud processing. Worrying about where their voice data was going. So I turned it into a product.
What Voiced is (and isn't)
Voiced is a local voice-to-text app for Mac. It does one thing: you hold a key, you talk, you release, and clean text appears wherever your cursor is. That's it.
It's not an AI writing assistant. It doesn't rewrite your words or change your tone. It doesn't do meeting transcription or subtitle generation. It's dictation — fast, private, and simple.
- 100% on-device. Your audio never leaves your Mac. Not optionally, not with a toggle. It physically can't.
- No subscription. $40 once. I got tired of paying monthly for cloud tools, so I wasn't going to do that to anyone else.
- No setup. No model selection, no API keys, no account creation. Download, grant two permissions, start talking.
- Works everywhere. Mail, Slack, Notes, VS Code, terminal apps, AI coding tools — if there's a cursor, Voiced works there.
I built the tool I wished existed when Willow started having problems. If you've ever been frustrated by cloud latency, subscription fatigue, or just the feeling that your voice data shouldn't be on someone else's server — that's exactly why Voiced exists.