Speak, and it’s written.Select, and it’s read.
Your voice, typed into any app you are using. Your content, read back in a human-like voice. Runs locally or in the cloud. Your software, your choice.
Free. Runs offline. Open source soon. Windows, macOS and Linux.
Works everywhere you write
Dictate into any app
Hold or tap your shortcut, say what you mean, and let go. The words land where your cursor was. A small bar shows it is listening. Nothing else on your screen changes.
- Whisper, on your machine. A small model that answers in a blink, or a larger one that knows a hundred languages.
- Terminals and stubborn apps. A paste that works in PowerShell; letter‑by‑letter typing where pasting is blocked.
- Cloud engines, if you want them. Off until you add a key, and then only for the parts you choose.
It knows where you are
The clean-up reads the room. In an editor it sees the open file and writes names the way the file spells them, whether you are asking the assistant or writing a comment. In mail it writes sentences. In a terminal it keeps your flags.
- Rules first, on this computer. Enough for most sentences, and instant.
- A model when you want more. Local, or one you bring a key for. Your tone stays yours.
- Your words, your spellings. A wordbook for names it should never get wrong.
Corrections, out loud
um so can we move the thursday review to friday at three no wait at four
Can we move the Thursday review to Friday at 4?
Formatting, said
todo colon new line bullet fix the login page new line bullet ship the installer friday
TODO:
• Fix the login page
• Ship the installer Friday
Names, numbers, dates
send david the q three invoice for two thousand four hundred euros by friday the fifteenth at three pm
Send David the Q3 invoice for €2,400 by Friday the 15th at 3 pm.
Clean read: the words, without the sidebar, buttons or timestamps.
Launch checklist
Ship the installer on Thursday once legal signs off. David owns the numbers page and Sam owns the announcement. If the sign-off slips we move the date and say so early. Let's make this one count!
The test expects null for a user that does not exist but the query throws instead. Return the row or null rather than letting the driver raise.
- return await db.users.get(id); + const row = await db.users.find({ id }); + return row ?? null;Code block skipped
With that change, the test passes and callers keep the contract they had.
Hi both, attaching the Q3 numbers. Revenue is up 12% on Q2, mostly from the two enterprise renewals, and churn is flat. The one thing I want eyes on is the support cost line, which grew faster than revenue again. Can we meet this Friday?
Q3 revenue is up 12 percent and churn is flat, support cost is growing faster than revenue. David asked: Can we meet this Friday?
Press play to hear Heart, the natural voice, read this one. Muted until you do.
Hear any text read aloud
Select text anywhere and press your reading shortcut. A natural voice reads from your selection, or the whole window if nothing is selected. Listen while you make coffee, walk to the window, or just rest your eyes for a page.
- Reads what matters. A clean read skips menus and code. Or read everything. Or ask for the summary.
- Skip, speed, pause. Ten seconds back, a little faster when you want, and it keeps its place while you answer a message.
- A voice that runs locally. Or Windows’ own voices, with nothing to download.
How fast do you type, honestly?
Type this exactly, capitals and commas included. Mistakes count even when you fix them, and it only ends when every character is right. Then we do the maths.
Your setup, in one sentence
Settings are written as plain sentences. Every green word is a choice: tap one to change the microphone, the language, the model, or how much clean-up you want. Try it here; this is the app, page by page.
Your setup
Tap any green word to change it.
Dictation
Nothing you say leaves this computer.
Read aloud
When I press CtrlAltR, it reads what I’ve selected, or the whole window, in Heart’s voice at 1.2 times normal speed.
Recent activity
This month
You saved about 1 h 40 min by speaking instead of typing.
Each day
Where you dictate
Dictation
How it listens, and what it does with the words.
It listens through my default microphone and understands English with the Base model, on this computer.
Then it cleans it up, keeping my tone and pastes it into the app I’m using.
While I speak, the bar shows the words as they arrive, at the bottom of the screen.
Read aloud
What it reads, and how.
When I press CtrlAltR, it reads what I’ve selected, or the whole window, in Heart’s voice at normal speed.
It reads just the words, skipping code and menus, and it turns other sound down while it speaks.
Voices on this computer
Words
What it should write when you say something.
Names and spellings
Words it should never get wrong, however they sound.
Replacements
Say the short thing, get the long one. Dates are worked out on the day.
History
Everything dictated or read, with its recording. Yours to replay, copy or delete.
Today
Settings
The app itself.
DualWhisper starts when I sign in and lives in the tray.
It looks the way Windows does, with green for the words I can change.
My shortcuts are CtrlAltD to dictate and CtrlAltR to read.
Private by default
Most dictation apps send your voice to a server and keep a copy. This one has no server to send it to.
Your voice stays here. Recognition, clean-up and the natural voice run on your own machine. Nothing is sent anywhere unless you add a cloud key yourself.
Your history is a file you own. Every dictation is kept locally with its recording. Replay it, copy it again, or delete it.
Nothing to sign up for. No account, no subscription, no telemetry.
And a command line
The same engine from a terminal. Watch it run, then type a command of your own.
Questions, answered
The short version. Everything else is in the setup, one sentence at a time.
Does DualWhisper work offline?
Yes. Speech recognition, the clean-up and the reading voice all run on your computer. Cloud engines from Groq or OpenAI are optional, for when you want the fastest transcription and do not mind sending audio.
Which apps does dictation work in?
Any app with a text field: email, Slack, Notion, a code editor, a terminal, a browser. Press the shortcut, talk, and the words are typed where your cursor is.
Can it read any text aloud?
Select text anywhere on Windows and press the read-aloud shortcut. A natural voice reads it at the speed you choose. In a chat with an AI assistant it reads the answer and skips the code.
Which languages does it understand?
The Whisper speech models understand English, French, German, Spanish and dozens more. Pick a language in setup, or let it detect the one you are speaking.
Is there an account or a subscription?
No. Download the installer, or install it with winget, and start talking. Nothing is uploaded unless you choose a cloud engine.
Does it run on macOS or Linux?
Not yet. DualWhisper runs on Windows 10 and 11 today. macOS is next, then Linux.
Get DualWhisper for Windows
Windows 10 and 11. No account, no subscription. Install, press the shortcut, start talking.
- WindowsAvailable now
- macOSComing next
- LinuxComing next
