Secure dictation app for business, on-device. Softspoken.

Where dictation usually goes

There are three common routes from your voice to the page. Which one a dictation tool takes decides where your words go.

A cloud dictation tool

The speech is sent to a server and turned into text there, by the vendor or by a speech-to-text provider it uses. Wispr Flow and Fireflies Talk work this way. So does the voice typing built into Windows, which uses Microsoft's online speech recognition and needs an internet connection. Some of these tools say they delete the audio once it has been transcribed.

A dictation system built for firms

Some secure dictation software made for medical and legal practices records the dictation on a phone or a computer and sends it on to be typed: by the firm's own typists, by an outside transcription service, or by speech recognition. The recording goes to whoever turns it into text.

On-device dictation

The speech becomes text on the computer you are using, so the audio has nowhere it needs to go. A Mac's own dictation can work this way, and Apple's guide says its Keyboard settings show whether yours does. Some dedicated dictation apps transcribe on the device too, and so does Softspoken.

Tidying the words is common ground as well. On recent hardware both operating systems now clean up a dictation on the device. With Apple Intelligence in macOS 27, a Mac with M3 or later improves spelling, punctuation and capitalisation in English. On Copilot+ PCs, Windows' Fluid dictation corrects grammar, punctuation and filler words.

Why not your computer's own dictation?

For many people it is enough. It comes with the computer, it types into any app, and on recent hardware it tidies the words on the device. On a Mac it can also turn the speech into text on the device, and the Keyboard settings say whether yours does.

Where a dictation app earns its place is the cleanup: what lands at your cursor reads as something you wrote, not a record of how you talked. Softspoken's cleanup takes out the ums, the false starts and the repeated words, fixes the punctuation and the run-on sentences, and keeps your words and your meaning. Then it writes to your settings.

Tone. Keep your wording as you said it, or make it more casual or more professional.
Email layout. When a dictation reads as an email, it is laid out as one: the greeting and the sign-off on their own lines, and lists as bullet points.
Spelling. Australian, British or American English, with Australian the default.
Style notes. Rules of your own, such as numbers as digits, applied wherever they do not change what you said.
Your Dictionary. The names and terms of your work, kept on your computer, come out spelled right.
Anti-AI writing. An optional setting takes out phrasing that reads as machine-written.

On a Mac, Apple's Writing Tools can rewrite text you select in a friendly, professional or concise tone, or in a way you describe. That is a second step, once the words are on the page. With Softspoken's cleanup on, the dictation arrives already written the way you set it, in any app.

On Windows, voice typing sends your speech to Microsoft's online speech recognition. Softspoken transcribes on the computer on Windows as it does on a Mac, so a firm with both gets one dictation tool that behaves the same way on each, with or without a connection.

The same app records and transcribes your meetings on the computer, so your dictation and your meeting notes live in one place.

If you dictate short messages on a Mac whose Keyboard settings show dictation runs on the device, and the words as you said them, tidied, are all you need, the built-in tool may be enough.

Six questions to ask any dictation app

On a product page, secure is a label. These six questions turn it into answers you can check. Put them to any dictation tool, ours included.

Where is the speech turned into text? On your computer, or on a server. If it is a server, ask whose: the vendor's, or a speech-to-text provider's that the audio is passed to.
Does it work with no connection? A tool that transcribes on the computer can. Some tools record with no connection and transcribe once you are back online, and then the audio still goes to a server.
Is the audio kept, where, and for how long? On your computer or on a server; deleted once it is transcribed, after a set period, or kept until you delete it. Ask whether you can change that.
Is anything you say used to train a model? Never, only if you opt in, or unless you opt out. For a firm, ask whether the setting can be locked for everyone.
What does the AI step see? If a tool uses a language model to tidy the words, ask what it sends: the text or the audio, and whether it also reads the app you are typing into to format the words for it. Some do.
Will your IT team say yes? Ask whether the vendor says plainly what the app connects to and how it is installed. That is what lets the person who approves software at your firm agree to it quickly.

Softspoken, Wispr Flow and Fireflies Talk

Wispr Flow and Fireflies Talk both offer dictation in any app beside meeting notes, as Softspoken does, so the three sit side by side.

	softspoken.	Wispr Flow	Fireflies
Dictation included	Yes	Yes	Yes
Meeting notes included	Yes	Yes	Yes
Audio never leaves your device	Yes	No	No
Works offline	Yes	No	No

How we checked

Their two columns were read on each vendor's own website on 27 October 2026. No cell is Partly, which would mean a phone app, separate hardware, a single platform or a separate mode.

Where they are level

All three put dictation in any app, on Mac and on Windows, keep meeting notes in the same app, and clean up what you say into finished writing. Wispr Flow formats the words for the app you are in and can apply a tone or format you choose, Fireflies Talk can also improve text you have already written on a Mac, and Softspoken writes to the tone, layout and rules you set. On the audio afterwards, Wispr Flow says it deletes it once it has been transcribed, unless you turn on its cloud storage for dictation. Fireflies says it keeps no copy of your dictations, and that its providers keep nothing once they have processed them.

Where they differ

Wispr Flow and Fireflies Talk send the speech away to be transcribed: Wispr Flow to its own servers, Fireflies to the speech-to-text providers it uses. So neither transcribes with no connection. Softspoken transcribes on your computer, with a connection or without one. Its dictation needs nothing first, and its speech model for meetings is a one-time download.

Both also read some text from the app you are typing into, to format the words. Wispr Flow always uses the app's name and the text around the cursor, and lets you turn off its wider reading of the window. Fireflies reads what it calls screen context, which you can turn off. Softspoken's cleanup sends nothing from that app.

Our guide to private AI notetakers sets the meeting side of both beside Softspoken.

Dictation that transcribes on your computer

Softspoken is a dictation tool for Mac and Windows as well as a meeting notetaker, and its dictation works the same way on both. Wherever you can type, you can hold the dictation key and speak instead. A speech model on your computer does the transcription, and the audio is never uploaded.

That makes it an offline dictation app from the first launch. The speech model ships inside it, so there is nothing to download first, and dictation carries on with every connection blocked.

Setup starts with dictation: you hold the hotkey, say a sentence and see it appear, while the speech model for meetings downloads in the background.

By default each dictation's recording is deleted as soon as its text is saved, and the text stays on your computer for seven days. Settings can keep the text for 30 days or for good.

Your meetings are recorded and transcribed in the same app, in the same way. Our guide to on-device transcription explains how that works for a meeting.

You need a Mac with Apple silicon on macOS 15 or later, or a 64-bit Windows computer. Softspoken is in closed beta, and the waitlist is where invitations come from.

What cleanup sends, and where it goes

Cleanup is the one part of Softspoken's dictation that uses a language model, so it is the one part that sends anything, and it is optional. It sends text, never audio.

What it sends is the dictation's text, with the Dictionary terms you have added and your style settings: the tone, the email layout, the spelling and your style notes. Nothing from the app you are typing into goes with it.

You choose where the text goes.

Softspoken AI. Your dictation is cleaned up in Australia and sent straight back. It is not stored, logged or used to train anything.
Your own account. On a paid plan, the AI steps can run on your own Anthropic or AWS account, from meeting summaries and questions across your meetings to this cleanup. The text goes straight to that provider, never through us. With Anthropic it will usually be processed outside Australia; with AWS, it is processed in Australia.
Off. Dictation still works, and the words land without cleanup.

How it works has a diagram of what leaves your computer under each choice.

Secure dictation app for business: what secure has to mean