Speakmac
The offline dictation macOS should have had
Start new thread
Kiran Jd

Speakmac - Private voice typing for Mac that works anywhere

by
Speakmac
•
Speakmac is private voice typing for Mac that works anywhere. Dictate with live preview, hands-free mode, custom words, multilingual packs, and audio-file transcription. It runs locally, feels fast, and gives Mac users a one-time purchase instead of another subscription.

Add a comment

Replies

Best
Kiran Jd
Maker
📌
Hey PH — I’m Kiran, the maker of Speakmac. Since the last launch in November, feedback from 170 users taught me that privacy alone isn’t enough — a dictation app has to feel instant, work in real apps, and be reliable enough to trust every day. As more of my work has shifted toward AI, I’ve found myself dictating more and typing less. Speaking gives me more control, more context, and a much faster way to get ideas into the tools I use every day. Speakmac is now a much more complete private voice-typing tool for Mac, with live preview, hands-free mode, custom words and regex replacements, better support for apps like Notion and VS Code, multilingual packs including Chinese, audio-file transcription, and a lot of reliability work behind the scenes. The core idea is still the same: your voice stays local, it stays fast on your Mac, and you buy it once instead of subscribing forever. I built Speakmac because I wanted the dictation tool macOS should have had. Would love to hear what you think :)
Tal Elor

Nice direction! voice is still one of the most underused “input layers” on desktop, especially when it actually works everywhere without breaking flow.

Philipp Kundratitz

Nice, I'm actually looking for a solution like that. Whats not clear to me from the landing page is what model it uses? I started trying Superwhisper, which had me install a local Parakeet model. Does yours do something similar, or how does it work behind the scenes?

Kiran Jd
Maker

@pkundr Speakmac uses Parakeet too as default(for now). There is now a more accurate model to comparable speed that you can use daily. You can download this model as add an add-on. I have been finding it gets the last 10% of words that Parakeet usually misses.

How is works: Press a hotkey to record audio from mic -> once done, calls the on-device model to transcribe the full audio file at once(Realtime gets too unreliable) -> text gets pasted in

Vijay Navaluri

Offline dictation is underrated. For founders/operators, that matters. Half the value of voice is capturing messy thoughts before they get cleaned up too early.

Στεφανος Βουδουρης

Hands-free with live preview is the interesting part, since the preview path and the final decode usually have different latency profiles and correction overhead. How does the system behave on code switching utterances, if someone dictates English and drops a Greek or Chinese proper noun mid sentence, does the multilingual path transcribe both correctly or does it commit to one language at the session level?

Kiran Jd
Maker

@myultidev Very hard for the models to handle 2 languages in the same session.
It tries to infer which language in the first few seconds of the clip.

There is another smarter model that needs you to select which language you're going to speak in.