Privacy by design
There is no server that receives your dictations. What is never sent cannot leak.
100% local · GDPR-compliant without the cloud
About SpeechToWork: made in Magdeburg, running only on your computer
About SpeechToWork in one sentence: it is developed by À Tellinghusen Software in Magdeburg, Germany, owned by Danilo à Tellinghusen. The dictation software runs entirely on your device, on Windows and on the Mac. This page explains who is behind it, why nothing goes to a cloud and which technology we use.
- 100%local: dictations, texts and receipts stay on your computer.
- 2 systemsWindows 10/11 and Mac with Apple silicon, same features.
- 12specialist dictionaries for German professional language, from radiology to tax.
An example
What you say. What appears.
You sayPlease transfer the amount by the fifteenth, no sorry, by the twentieth of October to the account shown on the invoice.
In your programPlease transfer the amount by 20 October to the account shown on the invoice.
Who is behind SpeechToWork
I am Danilo à Tellinghusen and I develop SpeechToWork in Magdeburg, Germany. I read and answer questions, bug reports and requests myself.
SpeechToWork is made by À Tellinghusen Software, owner Danilo à Tellinghusen, Breiter Weg 10a, 39104 Magdeburg, Germany. The software is developed, maintained and supported from there. If you have a question, want to report a bug or would like a feature, write to kontakt@speechtowork.de.
SpeechToWork is built for people who write a lot and handle confidential information: offices, law firms, medical practices, accountants, trades businesses and estate agents. For them, dictating should become as natural as typing, without client data ever leaving the building.
Why SpeechToWork runs locally
When you dictate, you talk about clients, patients and colleagues. Those words do not belong on a service provider's server. That is why privacy in SpeechToWork is not a setting but the way it is built: speech recognition and the language model are files on your computer and run there.
- Your dictations, texts, recordings and receipts are not sent to us or to any AI provider.
- SpeechToWork only goes online to download the models once (4 to 6 GB depending on your computer) and to check your licence.
- Once installed, dictation, rewriting, summaries and minutes also work without internet.
What this means in terms of data protection is explained in our guide to GDPR-compliant AI. Which data is transferred when you start the trial is set out in our privacy policy.
The technology, named openly
We say openly what SpeechToWork works with. All three building blocks are publicly available and run on your own hardware:
| Component | Job | Origin |
|---|---|---|
| Parakeet TDT 0.6B v3 | Speech recognition, detects 25 European languages automatically | Speech model by NVIDIA with openly available weights |
| Gemma 4 (E2B or E4B) | Language model: tidies up, rewrites, summarises, reads receipts | Model family by Google with openly available weights |
| llama.cpp | runs the language model on your computer | open source project, with Vulkan on Windows and Metal on the Mac |
On start, SpeechToWork checks memory and graphics and picks the right model size. The models are downloaded once as files and not changed after that. Your content is never used for training. More on this in our guide to local AI.
Windows and Mac, same features
SpeechToWork is available for 64-bit Windows 10 and 11 and for Macs with Apple silicon (M1 or later) running macOS 14 or later. Both versions do the same: dictation into any program, finished emails in your own style, editing selected text by voice, call notes, meeting minutes without a bot, summarising and asking questions about documents and audio, a Memory for details you use again and again, and reading receipts. 16 GB of memory is recommended. Details on the Mac version are on the page dictation software for Mac.
Pricing without small print
There are two plans, both per user and plus VAT: Dictate at 19 € a month or 192 € a year, and Assistant at 39 € a month or 384 € a year. You can try every feature for 14 days without payment details, and the trial ends on its own. Specialist dictionaries for German professional language cost 39.99 € each, once. All plans are compared under pricing.
Contact
À Tellinghusen Software
Owner: Danilo à Tellinghusen
Breiter Weg 10a
39104 Magdeburg
Germany
Email: kontakt@speechtowork.de
All mandatory details are in the legal notice.
Your advantage
Why SpeechToWork fits this job.
Open technology
Parakeet, Gemma and llama.cpp are publicly documented. You know what runs on your computer.
A real contact in Germany
Developed and supported by À Tellinghusen Software. Questions go straight to kontakt@speechtowork.de.
Privacy
Local means local. No compromises.
Speech recognition and the language model are installed on your computer and run there. What you dictate ends up in your program and nowhere else. How the data flow works.
No upload
Audio and text stay on your computer. There is no server listening in and no AI provider in the background.
GDPR made simple
No third party processes your dictations. So there is no data processing agreement to sign and no international transfer to assess.
Works without internet
Once installed, SpeechToWork works offline. Only the licence check needs a connection.
Good to know
Frequently asked questions
Who is behind SpeechToWork?
SpeechToWork is developed by À Tellinghusen Software, owned by Danilo à Tellinghusen. The business is based at Breiter Weg 10a, 39104 Magdeburg, Germany. The dictation software for Windows and Mac is built, maintained and supported from there. Questions, bug reports and feature requests reach us at kontakt@speechtowork.de.
Which AI models does SpeechToWork use?
For speech recognition, SpeechToWork uses Parakeet TDT 0.6B v3 by NVIDIA, which recognises 25 European languages. Rewriting is done by Gemma 4 from Google in the E2B or E4B size, depending on your computer. The language model runs with the open source llama.cpp. All three run locally on your device.
Can À Tellinghusen Software see my dictations?
No. Dictations, texts, recordings and receipts are processed on your computer and are not sent to us. There is simply no technical route by which they could reach us. SpeechToWork only needs a connection to download the models once and to check your licence, and no content is sent in either case.
Why does SpeechToWork use local AI instead of the cloud?
Because almost every dictation at work contains personal or confidential information. With local AI there is no data processor, no international transfer and no provider terms that can change overnight. The trade-off is a smaller model than in large data centres, but it is well suited to everyday writing work.
Is SpeechToWork available for Windows and Mac?
Yes. SpeechToWork runs on 64-bit Windows 10 and 11 and on Macs with Apple silicon (M1 or later) running macOS 14 or later. The features are the same on both systems, and so are the prices and the trial. 16 GB of memory is recommended. The Mac version is notarised by Apple and installs like any other app.
How do I contact the developer?
The quickest way is email to kontakt@speechtowork.de. The postal address is À Tellinghusen Software, owner Danilo à Tellinghusen, Breiter Weg 10a, 39104 Magdeburg, Germany. All mandatory details are in the legal notice. For questions about licences, invoices or features, it helps to mention whether you use Windows or a Mac.
Read on
You might also find these useful.
Local AI
What local AI is, which hardware it needs and what it does in the office.
ViewMac dictation
Dictate into Mail, Outlook, Word, Safari and any other Mac app, processed locally on Apple silicon.
ViewDictation software compared
Dragon, Windows voice typing, Apple Dictation, Wispr Flow, Superwhisper, SpeechLive, ChatGPT and SpeechToWork, with sources.
ViewFrom €16 net per seat per month · 14 days free trial, no payment details · For Windows and Mac Compare prices
Try it now
Download now and try it for 14 days.
One click, one file, no form. Install SpeechToWork and start the trial right in the app.
Windows: Version 1.0.0 · 87 MB · Windows 10/11 (64-bit), min. 8 GB RAM and 10 GB free
Mac: Version 1.0.0 · 87 MB · macOS 14 or later, Apple silicon (M1 or later), min. 8 GB RAM and 10 GB free
Download warning? Here is why.
SpeechToWork is new. Chrome, Edge and Windows SmartScreen only trust a file once it has been downloaded many times. For a new program they therefore show “Suspicious download” or “Windows protected your PC”. That is a judgement based on how well known the file is, not a malware finding.
- In your browser: keep the file in the download list.
- On Windows: “More info”, then “Run anyway”.
The installer is digitally signed, publisher Danilo à Tellinghusen. To check, right-click the file, choose “Properties” and open the “Digital Signatures” tab. The Mac version is notarised by Apple.
- All features, no payment details
- Ends automatically, nothing to cancel
- Your dictations never leave your computer, not even in the trial
Download
One file for Windows or Mac, straight from our server.
Install
On Windows a double click is all it takes, on a Mac you drag the app into Applications. No administrator rights needed.
Choose “Try for 14 days”
On first start: enter your name and business email, no payment method.
On first start, SpeechToWork downloads the language models once (4 to 6 GB). Already have a licence key? Enter it on first start. What is transferred in the process is explained in our privacy policy.