Your text, your way
Speakybara is a free app for Windows and Mac that turns any part of your screen into text you can copy or listen to: press a hotkey, drag a frame, done.
Everything below is grouped by what you actually do with it: put a frame around a piece of your screen, put the text on your clipboard, and listen to it.
Drag a frame around any part of your screen
Press the hotkey and the screen dims. Drag a frame around the text you want; dashed guides and a size chip show exactly what you are grabbing. When you let go the frame stays, with a small bar under it: Enter does what the hotkey says, the bar has the other choices, Esc cancels. More text than fits on the screen? Scroll inside the frame and the rest comes along. Turn off Frame stays after dragging under Settings › Capture if you want it to start the moment you let go.
It reads pixels, so it works on
- a scanned PDF with no text layer
- an image inside an email
- the screenshot a colleague pasted in Teams
- an error dialog you cannot select from
- a dashboard, a chart label, a map
- a remote desktop or Citrix window
- a paused video
Not a screenshot manager
Speakybara has no annotation tools, no uploads and no share links, and its history keeps only the text and a small preview, on your own computer. If you already use CleanShot X, ShareX or the Snipping Tool, keep it. Speakybara does the part they leave out: getting the text out of the picture.
The recognised text on your clipboard, in one press
The same frame, a different hotkey: press Enter, the text lands on your clipboard and you paste it where you were going to type it anyway.
Recognised on your own computer
The text recognition that ships with Windows or macOS does the work, on your machine; on Windows that takes 50-150 ms. The image goes straight into it and is discarded afterwards; the recognised text is kept in a local history you can switch off, and by default no image leaves your computer.
Two recognisers, best result wins
On Windows, Speakybara reads your selection twice, once with the Dutch recogniser and once with the English one, and keeps the result with the most real words in it. In the settings you can also pin the recognition to one language.
Enlarge for small type
Small type on a busy screen is recognised better when the image is enlarged first. The slider runs from 1.0x to 4.0x and sits at 2.0x by default: higher is better on small letters and slightly slower.
Keep the shot if you want it
The frame you dragged can be saved as an image file. That is a convenience, not what the app is for.
A small window that reads along with you
While Speakybara speaks, a small dark window shows the text. The sentence being spoken has a blue background and the words turn white as they are said, so you can look away and still find your place.
What the window does
- The sentence being spoken has a blue background.
- Words that have been said turn white.
- Click a sentence to continue from there.
- Drag the top bar to move it, the corner to resize it.
- Position and size are remembered.
- You can switch the window off entirely in the settings.
Speed
Seven steps, from 0.8× to 1.75×. The step you pick stays until you change it.
- 0.8×
- 0.9×
- 1×
- 1.1×
- 1.25×
- 1.5×
- 1.75×
While it is reading
- CtrlAltPPause or resume
- CtrlAltXStop
- CtrlAltRRead the last text again
Explain, summarise, translate
A capture does not have to end on your clipboard. The recognised text can go straight into an action: from Quick action in the bar under the frame, from the ⋮ menu on Home, from the Continue with panel after a capture, or from the reader's own ⋮ menu. Ctrl+Alt+Q always keeps the frame and its bar, even with Frame stays after dragging switched off.
Explain — when something does not make sense
The error screenshot a colleague pasted into Teams, or a paragraph of jargon halfway down a report: Explain with AI answers in plain language, and reads that answer out the same way as any other text.
Summarize — long or dense text
The forty-line mail at half past four, or a page you have no time for: Summarize gives you the gist first, so you can decide whether the rest is worth reading.
Translate and improve writing
Translate reads a capture out in the language you pick. Improve writing tidies up your own draft and, unlike the other actions, always puts the result on your clipboard, ready to paste back.
Your own actions, up to eight
A name, an instruction for the AI and one of 24 icons is enough. Your action then stands next to Explain and Summarize everywhere they already are: the ⋮ menu on Home, the reader, History, and Quick action in the bar under the frame.
Your own key or a local model
Speakybara runs no AI service of its own. Actions run on an OpenAI key you paste in under Models › AI model, or on a model you run yourself — LM Studio, Ollama, llama.cpp or Unsloth/vLLM — with no key and no internet.
Four voice providers, and what each one costs you
Piper, the voices of Windows or macOS, Microsoft Edge (on Windows) and Google Cloud offer over 100 voices across languages. Availability varies by language, provider, platform and installed voices. None of the four uses your graphics card.
| Service | Quality | Cost | Internet | VRAM | Example voices |
|---|---|---|---|---|---|
| Local (Piper)FreeOffline | Fair | Free | no | 0 MB | nl_NL-ronnie-medium · en_US-amy-medium |
| Windows or macOS voicesFreeOffline | Varies by installed voice | Free | no | 0 MB | Microsoft David · Microsoft Zira · Xander (macOS) |
| Microsoft EdgeFreeWindows only | Good | Free, no limit | yes | 0 MB | nl-NL-FennaNeural · en-US-EmmaNeural |
| Google Cloud | Best | Free allowance with your own key, or Complete | yes | 0 MB | nl-NL-Chirp3-HD-Achernar · en-US-Chirp3-HD-Achernar |
Free allowances, what happens offline, and the usage meter
What Google gives away
With your own key, Google's free allowance is 4,000,000 characters a month for Standard and WaveNet and 1,000,000 characters a month for Neural2 and Chirp 3 HD. A page of running text is about 3,000 characters. The meter in the settings counts each family separately, because the limits differ.
When the internet drops
If an online service becomes unreachable, Speakybara carries on with the local voice and says so once — not once per sentence. When the service comes back, it picks it up again.
A meter, and a cache for sentences you already heard
Online voices cost characters, so the app counts them where you can see them.
- The meter shows characters used this month per voice family, what share of the free allowance that is, and roughly how many pages are left.
- Every sentence spoken by an online voice, such as Edge or Google, is stored on your computer. Read the same thing again and it starts immediately and costs no characters.
- The cache stops at 600 files or 250 MB and then drops the oldest.
- Google's own console shows requests, not characters, and stays at zero while you are inside the free allowance — which is why Speakybara keeps its own count.
35 languages, with a voice for each
The app itself is translated into all 35, and it follows the language preferences of Windows or macOS. Choose active languages and a voice per language. Many languages can be detected automatically; others can be selected manually. A few, such as Hebrew and Hindi, cannot be read from the screen by Windows or macOS: copy the text and have the clipboard read aloud.
Per selection
Windows recognises text using your installed OCR language packs; the Mac has its languages on board. Choose the languages you want Speakybara to use.
Per sentence
When a language is detected, Speakybara selects the voice you assigned to it. You can also set a language manually.
Per word
On Windows, an English term in the middle of a Dutch sentence is pronounced in English by the same Piper voice, so the sentence keeps one voice and one melody. The Mac version does not do this.
English words in a Dutch sentence, and your own jargon
English words in a Dutch sentence
A voice reads sound symbols, not letters. Speakybara takes the sounds of English words from the English rulebook and the rest from the Dutch one, and hands both to the same speaker. This applies to Piper on Windows only: the Edge and Google voices already read English words well, so the option is off for them.
| Word | Without | With |
|---|---|---|
| pull request | pul rayk-vest | pull rikwest |
| deadline | day-ahd-lee-nuh | dedlain |
| pipeline | pee-puh-lee-nuh | paipline |
Dutch words that happen to look English stay Dutch: brand, post, model, test, camera, export.
Your own jargon
Term not in the list? Put it in english-words.txt, one per line. The file is re-read while the app runs, so nothing has to restart.
Hotkeys you set yourself
Assign your own shortcut to captures, clipboard reading and custom actions. These are the main defaults.
| Action | Keys |
|---|---|
| Frame a region and read it aloud | CtrlAltS |
| Frame a region and copy the text | CtrlAltE |
| Frame a region and run the AI action set in Settings, then read the result aloud | CtrlAltU |
| Capture and choose an action | CtrlAltQ |
| Read the clipboard aloud | CtrlAltC |
| Pause or resume | CtrlAltP |
Ten seconds back or forward, and the previous or next sentence, are buttons in the read-along window.
Changing a hotkey
Changing one takes four seconds
Click a hotkey field in the settings. It turns blue and counts down from eight. Press the keys you want and it is done — nothing to type, nothing to save. Esc stops it straight away. On Windows, Speakybara listens directly at Windows during those eight seconds, so combinations other programs normally swallow still arrive; those keystrokes go nowhere else.
- A combination without Ctrl, Alt, Shift or Win is refused.
- If another Speakybara action already uses it, it says so.
- If another program already holds it, it says so too.
What stays on your computer, and what does not
Text recognition is local. Your text stays on your computer with local voices and local AI. Cloud voices and online AI use your chosen provider; Complete routes voice text through the Speakybara service.
| What | On your computer | Sent away |
|---|---|---|
| The frame you dragged | Recognised locally; history and preview storage can be switched off | Nothing by default. The one exception is Explanation for a step, at the bottom of this table. |
| Local voices (Piper, Windows or macOS) | Speaks entirely on your computer | Nothing. Works with the network unplugged. |
| Microsoft Edge voices (Windows) | The image and the recognised text | The sentence being spoken goes to Microsoft. |
| Google voices (own key or Complete) | The image and the recognised text | The sentence goes to Google; with Complete it passes through the Speakybara service. |
| Your Google API key | Stored in config.json on your own computer | Only to Google, to authorise your own requests. |
| AI actions (Explain, Summarize, Translate, …) | With a local model, entirely on your computer | The recognised text, not the image, to the provider you chose. |
| Explanation for a step, with Explanations may look at your screen | Off by default. With a local model the picture stays on your computer too | A picture of your screen, to the online AI you chose, when you ask about a step |
An AI action — Explain with AI, Summarize, Translate, Improve writing, the optional AI clean-up, or an action you wrote yourself — sends the recognised text, and only the text, to the provider you chose yourself: an online service with your own key (OpenAI, Google Gemini or OpenRouter), or a model running on your own computer through LM Studio, Ollama, llama.cpp or Unsloth/vLLM. Your capture is not sent along, and with a local model nothing leaves your computer at all. Speakybara runs no AI service of its own and has no server in that path: the request goes straight from your computer to that provider, and your key stays in config.json on your own disk.
Small facts: memory, startup, tray and menu bar
Small facts
- On Windows, about 55 MB of memory in use, plus about 60 MB while a local Piper voice is loaded.
- On Windows, 0 MB of video memory, with all four voice providers.
- Text recognition is already part of Windows and macOS: no model to download.
- Pick a local voice and a 60 MB voice file loads. After 90 seconds of silence it is unloaded again; that time is adjustable.
- On Windows it lives in the system tray: double-click the icon to drag a frame, right-click for the menu. On a Mac it lives in the menu bar, with a Dock icon if you want one.
- Starting with Windows is a checkbox, and it only puts a shortcut in your Startup folder. On a Mac, opening at login uses the login items of macOS.
- Windows 10 (build 19041) or newer and Windows 11, 64-bit. On a Mac: macOS 14 or newer with Apple silicon (M1 or later), as a test version.
Also built for how you read
Listening instead of reading suits anyone now and then. It is worth more if reading costs you more than it costs other people.
When your eyes are done
Half past four, the mail is still forty lines long and the screen has been on since nine. Frame it, listen, and let your eyes do something else.
Read moreIf you're dyslexic
Listening moves the effort from decoding the letters to understanding the sentence. Speakybara lives in the tray or the menu bar and shows nothing on your screen until you ask it to.
Read moreIf you have ADHD
Hearing and seeing the same sentence gives your attention two things to hold on to. The read-along window keeps showing where you are, so coming back after a break is one glance.
Read moreReading in a second language
A Dutch letter today, an English report tomorrow, often inside one document. The language is detected per sentence, and on Windows even per word, so you switch nothing.
Literacy suites and screen readers
If you need more than listening
Read&Write, Kurzweil 3000 and ClaroRead are full literacy suites: reading, spelling, writing support, study tools, and often a route through your employer or school. Speakybara does one piece of that: the part of the screen you point at, copied or read aloud. If you need the rest, those do more than Speakybara ever will.
It is not a screen reader
NVDA, Windows Narrator and VoiceOver on the Mac drive the whole computer by voice, for people who are blind. Speakybara is a read-aloud tool: it reads the piece of screen you point at, including text inside images that a screen reader cannot reach. It sits next to them, it does not replace them.
Free and Complete are the same app
Everything above works in the free version. No locked buttons, no extra features behind a paywall, no watermark.
The same in both
All features, the local Piper voices, the Microsoft Edge voices on Windows, and Google's best voices — with your own key in Free, ready to go in Complete.
The only difference
Who sets up Google. In Free that is you, in about 15 minutes. In Complete it is us: €59 per year or €6.99 per month, incl. VAT.
What Free asks of you once
- Create a Google Cloud project
- Turn on the Text-to-Speech API
- Create an API key and restrict it
- Add a billing account
- Paste the key into the settings
Last updated September 12, 2026