Mr.Speech has a fully private mode that runs speech recognition, search and speaker recognition entirely on your own machine. When you do choose the cloud, it is processed in Europe. You pick which, per task.
Speech recognition, semantic search and meeting transcription all have full on-device implementations. No account, no connection.
When you choose accuracy over isolation, audio is processed on European infrastructure. The processors are named in our privacy policy.
The same recording, two paths. You choose per task, and nothing switches behind your back.
The same recording can be handled three different ways. Nothing switches behind your back.
The on-device speech engine handles 40 languages with real punctuation and runs about 7 to 9 times faster than real time on an ordinary processor. Semantic search and speaker recognition also have full on-device implementations. No account, no internet, no audio leaving the machine.
Speech recognition, document reading and AI summaries run on European infrastructure. The specific providers are named in our privacy policy so you can check them against your own procurement rules.
Point the AI features at your own provider account, an enterprise endpoint, or a model you run locally. You are never locked to one vendor, and neither is the product.
If you are paying for cloud transcription, Mr.Speech will not quietly swap you to a weaker engine when something fails. Errors are shown, not hidden. The only exception is running out of your monthly quota, and that is surfaced to you directly.
API keys and tokens are stored by the OS credential manager, never written into a config file next to your notes.
Your recordings, transcripts and notes are not a product we sell. Usage analytics can be switched off in settings.
To place dictated text correctly, Mr.Speech can read the field you are typing into. That is a setting you control, and it never touches password fields.
Meetings separate who spoke so a summary can attribute correctly. Turn labelling off and a meeting renders as You and Others only, with the two audio channels still kept apart.
Every note exports as PDF, as a self-contained HTML file, or as a folder of plain Markdown files that mirror themselves as you type. If you leave, you take everything, and it opens in any text editor.
Yes. Dictation, notes, semantic search and on-device meeting transcription all work offline. Cloud accuracy and the AI features are the parts that need a connection, and they are opt-in.
Recordings and notes live on your own machine. Meeting audio is written to disk continuously while recording, which is also what lets a call be recovered if the app or the computer stops unexpectedly.
The design intent is European first: a fully on-device path for work that cannot leave the building, and European processing for everything else. What you need for your own assessment, including the named processors, is in the privacy policy.
No. Your recordings, transcripts and notes are not used to train models, and they are not sold or shared for advertising.
No. There is no sentiment score, no mood detection and no speaker energy metric anywhere in the product. We explain why on the EU AI Act page.
Install it, turn the cloud off, and dictate for a week without a single byte leaving your computer.
