Mac app · 2026
Humdrum
I want meeting notes and dictation on my Mac without shipping my voice off to somebody's server. So I built Humdrum. Hit record and it writes a live transcript as people talk. Press ⌥Space in any app and whatever you say gets typed into the field you're in. Nothing leaves the machine.
- Pick a model and record. Gray italics are still being decided.
- ⌥Space in any app. Talk, pause, and it types for you.
- Speakers labeled after you stop. Save as .txt, .md, .rtf, .json.
It runs an open-source speech model, Whisper, converted to run on the Mac's own AI chip instead of a server. You pick a quality from Fast to Best, trading speed for accuracy. Words it's still unsure of show up gray and italic until they settle. When you stop, a second local model works out who was talking and labels the transcript Speaker 1, Speaker 2.
Feed the model silence and it gets creative. "Thanks for watching." "[BLANK_AUDIO]." It learned from a lot of captioned video. Humdrum's noise filter skips quiet audio before the model hears it and scrubs the leftovers after, so a real sentence that ends in "thanks" survives and a phantom sign-off doesn't.
The dictation orb floats over whatever you're typing in without taking over the window, so the paste lands where your cursor was. And on first launch it offers to move itself into Applications, because an app running from Downloads tends to lose its microphone and accessibility permissions the day you tidy up.
Next on the list is renaming Speaker 1 to an actual person, so a transcript reads like the meeting instead of a court record.
