Push-to-talk dictation that inserts in place: hold fn, speak, release, and the text appears in the app your cursor is already in. Transcription runs locally with whisper.cpp + Metal, so your audio never leaves the Mac. An optional on-device cleanup pass strips filler words and fixes punctuation. Most dictation apps show you a recording dot or a generic soundwave. This one has hand-animated indicators. $10 once, no subscription. MIT source. macOS 14+, Apple silicon.
Yo,
I write code all day and wanted dictation that could keep up with the words developers actually use.
Hold fn, speak, release. The finished text appears in whatever app your cursor is already in — editor, terminal, Slack, a commit message. Transcription is local whisper.cpp with Metal, so your audio never leaves the Mac.
The part I spent far too long on: every dictation app shows you a recording dot or a soundwave, and you look at that thing several times an hour. So I drew proper animated ones instead. Two ship with the app; three more are $3 each and purely cosmetic — they change nothing but what you look at while you talk.
There's also an optional local cleanup pass: a small on-device LLM that strips filler words and fixes punctuation without changing your meaning. Off by default.
Honest limits — macOS 14+, Apple silicon only, English only, and the first launch downloads a ~1.5GB model.
$10 once, no subscription, 14-day trial with no card. The source is MIT at github.com/WillisLiao/Phantom-Voice, because "your audio stays local" isn't worth much if you have to take my word for it.
Solo developer, first paid app. I'd genuinely like the feedback — including the unflattering kind.
the opt-in rewrite does more than save 1.1 gb — a cleanup model can turn a technical token it doesn't know into a fluent wrong one. keeping punctuation separate from semantic rewriting makes the risk visible.
Local plus push-to-talk is the combination I'd want — always-listening is the thing that makes people uneasy, and a held key is an unambiguous consent signal in a way that a wake word isn't.
The number I'd want to know is latency from key release to text appearing. Local models trade speed for privacy, and dictation is unusual in that a one-second delay is fine but a three-second one breaks your thinking — you've already moved on. Where does it land on an average Mac, and does model size let you trade accuracy for that if you need to?
Guys go download it on the site and try for free! I need feedback! Already found something that needs to be updated and that is the cleanuppass, now if toggled on, it uses a local LLM to clean up, but the downside is it requires another 1.1GB of RAM. A better way to clean up is what someone here pointed out: most of it is just string work. Fillers and stutters are a regex, and fixing my mis-transcribed technical terms is edit-distance matching against the
dictionary you already typed in - neither needs a model resident in RAM. The only part that genuinely wants an LLM is punctuation and casing, and whisper already does most of that. So the plan is to make the cheap path
the default and keep the 1.1 GB model as an opt-in for people who want the
heavier rewriting.
I've lost count of how times I've opened a diction app, waited for it to connect, and forgetten what I wanted to say.This push-to-talk workflow feels so much more natural. Great work!
Guys let me know if the indicator designs are anygood, or are there any improvements I can make with the app
About Phantom Voice on Product Hunt
“Local push-to-talk dictation for Mac”
Phantom Voice launched on Product Hunt on July 30th, 2026 and earned 84 upvotes and 15 comments, placing #21 on the daily leaderboard. Push-to-talk dictation that inserts in place: hold fn, speak, release, and the text appears in the app your cursor is already in. Transcription runs locally with whisper.cpp + Metal, so your audio never leaves the Mac. An optional on-device cleanup pass strips filler words and fixes punctuation. Most dictation apps show you a recording dot or a generic soundwave. This one has hand-animated indicators. $10 once, no subscription. MIT source. macOS 14+, Apple silicon.
Phantom Voice was featured in Mac (103.6k followers), Productivity (657.9k followers), GitHub (41.4k followers) and Menu Bar Apps (12.2k followers) on Product Hunt. Together, these topics include over 189.9k products, making this a competitive space to launch in.
Who hunted Phantom Voice?
Phantom Voice was hunted by Willis Liao. A “hunter” on Product Hunt is the community member who submits a product to the platform — uploading the images, the link, and tagging the makers behind it. Hunters typically write the first comment explaining why a product is worth attention, and their followers are notified the moment they post. Around 79% of featured launches on Product Hunt are self-hunted by their makers, but a well-known hunter still acts as a signal of quality to the rest of the community. See the full all-time top hunters leaderboard to discover who is shaping the Product Hunt ecosystem.
Want to see how Phantom Voice stacked up against nearby launches in real time? Check out the live launch dashboard for upvote speed charts, proximity comparisons, and more analytics.
Willis Liao