Private transcription software: Keep the audio on your device
Author: Mert S. Kaplan · Last updated: November 4, 2026
Private transcription software should not need your recording on someone else’s server. LiveType runs the Whisper model inside your browser, so with This device selected the audio never leaves it; only the resulting text and your account details reach our server. You do not have to take our word for it, and below we show how to check.
Where your audio goes: Three options
Where to transcribe? under Settings on the transcript page decides the route:
| Option | Where the audio is processed | Does it pass through our server? | When to use it |
|---|---|---|---|
| This device | In your browser | No | The default, when privacy comes first |
| Transcriber device | On another computer of yours or your team’s | No, it goes directly over an encrypted connection | When this device is too weak |
| Cloud transcription | At the provider you choose | Not with your own key; with credits it passes through unsaved | When speed matters most |
Files work the same way: to transcribe audio in the browser, the file is only opened there. That is why you are asked to choose it again after reloading the page. No other copy exists.
Try it on your own recording
No card needed. 30 minutes of transcription on the free plan.
What the server stores
Stored: the transcript’s text, title and versions, which sharing and access from other devices rely on; account details and time used; the names you give speakers.
Not stored: audio or video, as the server has nowhere to keep them; Whisper models, which your browser downloads directly from Hugging Face; voice prints, which stay encrypted on your device and are deleted after 60 days unused.
So the text does live on our server, and you should weigh that if the content itself is sensitive. The system administrator can also open a transcript’s text. Transcripts are deleted after a period that depends on the plan, and you can delete them any time. The details are in the privacy notice. In a sensitive conversation, a label like “Source 1” may serve better than a real name, since speaker names are part of the text.
Check it yourself in the Network tab
Your desktop browser’s developer tools list every request a page makes:
- Open a transcript page, press F12 and go to the Network tab.
- Make sure This device is selected, clear the list and press Start transcription.
- Let it run for a while, then read the addresses and the Size column.
| Request | What it is for |
|---|---|
huggingface.co, cdn.jsdelivr.net |
Downloading the model and runtime files, first use only, in the download direction |
/api/transcripts/…/usage, /screen |
Time used and which screen is open; very small |
/api/collab (WebSocket) |
Saving the text; its Messages tab shows frames about the size of text |
If you chose a 100 MB audio file, you will not find an outgoing request anywhere near that size. Tracking code added by the installation’s administrator would also show up in the list, and carries no audio. Repeat the test with cloud transcription and you will see the difference: pieces of audio going to an audio/transcriptions address.
When you choose the cloud
Cloud transcription spares your device and is very fast on files, but the audio leaves it. With your own API key, pieces go straight from your browser to the provider and its terms apply. With credits, they pass through our server without being stored on the way to the provider. For a confidential conversation, avoid both and look at the device field before you start.
All the options are compared under cloud transcription, credit and remote devices. For the legal side in Turkey, see KVKK and voice recordings.
What local transcription costs you
- Your device does the work. Without a graphics card you are on smaller models and accuracy drops; see browsers and devices, and our accuracy test for how much.
- The first download. A model can be hundreds of megabytes, once.
- It is not an offline transcription tool in every sense. Starting asks the server how much time you have left. After that, a dropped connection does not stop transcription, and the text is saved when the connection returns.
- Your room and computer are still yours to secure. Keeping audio on the device does not protect the recording itself.
In return you can edit while it transcribes, on audio nobody else receives: correct a name the moment you see it without sending the conversation anywhere. The free plan has 30 minutes in total, and tools that upload your files are set against this approach on the comparison pages.
Works in news, software and design; founder of LiveType. Worked as a reporter for many years. Has done SEO consulting, content editing and reporting for several online news sites, and has developed many pieces of software. Currently works in the press office of the Turkish Medical Association and provides SEO support to Elips Haber.
Related pages
Try it on your own recording
No card needed. 30 minutes of transcription on the free plan.