Skip to content
Privacy

Is online transcription private? What happens to your files

· 5 min read · Total Soft Solutions

Transcription feels harmless: you hand over a recording and get words back. But a recording can hold things you would never type into a public form, such as a client’s details, a patient’s symptoms, a source’s name, a family argument or a company’s plans. Where that file goes while it is being turned into text is a fair question.

This guide explains the two ways transcription tools work, what you should look for in any service’s privacy terms, and how to test a claim yourself in a few minutes. It does not tell you to avoid online services. It helps you match the tool to the sensitivity of the file.

Two ways a transcription tool can work

Almost every transcription tool falls into one of two groups. The difference is where the speech model runs.

  • Upload services. Your file is sent over the internet to the company’s servers. A model on their computers produces the text, and the transcript is sent back and often stored on their side too.
  • On-device tools. The speech model is downloaded to your device and runs there. The audio is read by your own browser or app, and only the model, not your file, crosses the network.

OmniaScribe is the second kind. The model is downloaded once, then reused, and your recordings and transcripts are processed and kept in your browser.

What “upload” can mean for your file

Uploading is not automatically unsafe. Many providers encrypt files in transit and at rest, delete them on a schedule and publish clear policies. But uploading does mean a copy of your recording exists on a computer you do not control, for as long as the provider keeps it. Questions worth asking include the following.

  • How long is the audio kept, and can you delete it yourself?
  • Is the transcript stored with your account, and who at the company can read it?
  • Does the provider use customer recordings to improve its own models?
  • Which other companies, such as cloud hosts or sub-processors, handle the file?
  • In which country is it stored, and what rules apply there?

The answers are usually in the privacy policy and the terms. If you cannot find them in plain language, treat that as an answer in itself.

Try it on your own file

or drop it here · MP3 WAV M4A MP4 MKV MOV FLAC OGG
Nothing is uploaded

When it matters most

For a podcast you are about to publish, upload risk is small. For other material, the stakes are real. Think about whether any of the following describes your file.

  • Interviews with sources or research participants who were promised confidentiality.
  • Client, patient or student conversations that carry duties of confidentiality.
  • Internal meetings where a leak would hurt your company.
  • Recordings that include other people who never agreed to be sent to a third party.

Professional rules in fields such as law, health care, journalism and research often limit where recordings may go. Your own rules or agreements come first, and this guide is not legal advice. What on-device tools give you is one simple fact to build on: the recording was never sent anywhere.

What stays on your device with OmniaScribe

Here is what the site does, as precisely as we can say it.

  • Your audio or video is opened by your browser and passed to a background worker. The speech model runs there, on your own processor.
  • The transcript, the title and the recording are saved in your browser’s own storage. You can delete everything from the Settings screen.
  • The only large download is the speech model, from a server that holds model files. It carries no information about your recordings.
  • The site has no account system and no database of transcripts, so there is nothing for us to hold, search or hand over.
  • Once the model is installed, you can turn off your connection and transcribe, which is a quick test of whether anything depends on a server.

No request the page makes ever carries your file or your transcript. The model server only sends model files to you, and nothing you transcribe is sent back to it.

How to check any tool yourself

You do not have to take anyone’s word for it, including ours. Two quick tests tell you a lot.

  1. Watch the network. Open your browser’s developer tools (usually F12, or the menu under More tools) and choose the Network tab. Transcribe a short file. An upload tool will show your file leaving in a large request. An on-device tool will show the model downloading the first time and no request carrying your file.
  2. Go offline. After the model is installed, turn off Wi-Fi or switch on airplane mode and transcribe again. If it still works, the work is happening on your device.

Run the same tests on a tool that uploads, and the difference is clear. The Network tab is also useful for a general rule of thumb: be wary of any private tool that cannot explain what it sends.

Choosing the right tool for the job

For confidential interviews, meetings and personal recordings, an on-device tool such as the audio to text page keeps the question of trust out of the picture. The same goes for the interview and meeting recording pages if that is what you are transcribing.

For public material where you want the very best accuracy and a human proofreader, a professional service that uploads may be the better fit, as long as you read its terms. Many people use both: on-device for the private files and a service for the public ones.

Accuracy is the honest trade-off. On-device models are smaller than the largest models running in a data center, and long files take longer on a phone than on a server. Read our guide to choosing a model size to see how to balance speed and accuracy on your own hardware.