Speech to text for Mac that stays on your Mac.
Two questions decide which one you want: does the audio leave the machine, and does it know the words your work uses. Everything else is detail.
The phrase covers two different jobs
Speech to text is used for two things that have almost nothing in common. One is transcription: you have a recording, an interview or a lecture, and you want it written out. The other is dictation: you have a cursor and a thought, and you want the words to appear where the cursor is while you talk.
Working out which one you need takes ten seconds and saves you buying the wrong thing. If the input is a file, you want a transcriber and MacWhisper is the usual answer. If the input is you, right now, you want dictation.
On the machine or on a server
This is the decision that sets everything else. If the model runs on your Mac, the text comes back in about a fifth of a second, it works on a plane, and nobody else ever holds a copy of what you said. If it runs on a server, you get looser hardware requirements and, usually, a monthly bill.
Zumbo ships the speech engine inside the download, which is why the download is about 570 MB. There is no account, no API key and no network request needed to transcribe. The only things it ever sends are your licence key when you activate and a version number once a day to check for updates.
Speed is a threshold, not a score
Hold it to talk, tap it to toggle, Escape cancels. Pick the key in Settings.
Under about a quarter of a second, the words feel like they came from you and you keep talking. Over it, you start waiting to see what appears, and once you are waiting you are dictating carefully, which is slower than typing. Zumbo composes locally in about a fifth of a second, on the right side of that line, and stays there with the network off because there is no network in the path.
Knowing your words
A general model spells what it thinks it heard, and what it thinks it heard is ordinary English. That is fine until your work has its own language, at which point every paragraph contains two or three words you have to go back and fix. Twenty one vocabulary packs cover fields like developer tools, legal, medical, design and marketing. For anything still wrong, correct it once with Teach and it is heard your way from then on, with the correction stored on your Mac.
What comes with it
Everything Zumbo writes stays in a searchable history on the Mac. You can press New note and talk without pasting anywhere, and a time you said becomes a reminder. Meeting mode transcribes the room through the microphone with a timestamp per paragraph and each speaker labelled, using an optional add on of about 220 MB for the labelling model.
What it costs
Fifteen dollars once for one Mac, twenty for three, seventy nine for ten. Free updates. Three day trial with everything unlocked and no card. macOS 14 or later on Apple silicon, because the engine runs on the Neural Engine.
Related: dictation with the Wi-Fi off, voice typing on the Mac, with the three year arithmetic, and all six apps side by side.
What is the difference between speech to text and dictation?
Speech to text is the general term. Dictation means the words arrive where your cursor is while you talk. Transcription means a recording you already have gets written out. The same phrase is used for both and they are different jobs.
Does speech to text work offline on a Mac?
It does if the model runs on your machine. Zumbo ships the speech engine inside the download, so there is no upload and no server to reach, and it behaves the same with the router unplugged.
How accurate is it?
On our own recordings, technical terms come out right 9 times out of 10 with the packs on, and every correction you make raises that for your voice. Normal sentences are where it shines; a single word chanted three times is where any engine struggles.
Why is the download 570 MB?
Because the speech engine is inside it. A small speech to text download almost always means the work is happening on somebody else's machine.
Does it handle more than English?
25 European languages, English first. You pick your main one during setup and it then writes only in that alphabet.
Free for 3 days, everything unlocked. Then $15 once. macOS 14 or later, Apple silicon.