Skip to content
Transcription

How to transcribe audio to text for free, without uploading it

· 6 min read · Total Soft Solutions

Most free transcription sites work the same way: you upload your recording, it waits in a queue on someone else’s computer, and a transcript comes back. That is fine for a podcast you are about to publish. It is a poor fit for a doctor’s dictation, a confidential interview, a family video or anything else you would rather keep to yourself.

This guide shows another way. OmniaScribe’s audio to text tool downloads a speech model into your browser once, then turns audio into text on your own device. Below you will find what you need, the steps from file to finished transcript, how to get a cleaner result, and what to do when a file is long.

What you need

A recent Chrome or Edge on a computer, a recording, and a little patience the first time. The first time you transcribe, OmniaScribe downloads a speech model to your device. After that, the model is stored in your browser and transcribing needs no connection at all.

The recording can be almost anything you already have. Audio formats include MP3, WAV, M4A, AAC, FLAC, OGG and OPUS, and a dedicated page covers MP3 to text if that is your format. Video files work as well (MP4, MOV, MKV and WEBM), because only the sound track is used. If you are not sure whether your file will open, just try it: the browser decodes the file, and if one method fails the site tries a second before it asks for a different format.

There is nothing to install and no account to create. There is no daily minute allowance either, so the only limit is how long your own device takes to work through a file.

From file to transcript, step by step

  1. Open the workspace with the “Start transcribing” button, or drop your file onto the card on any OmniaScribe page. The file is handed to the workspace in your browser, not sent anywhere.
  2. If this is your first time, download the model it recommends. The Models screen explains the three sizes, and our guide to choosing between Fast, Standard and High accuracy helps if you want to pick one yourself.
  3. Leave the language on automatic detection, or choose the language that is spoken. Choosing it yourself is more reliable when a recording starts with music, silence or a greeting in another language.
  4. Start, and watch the progress. The engine reads the sound in short windows and writes as it goes, so you can see how far along it is.
  5. When it finishes, the transcript opens with a player. Click a line to hear that moment, and use Edit to fix names or numbers.
  6. Export what you need. The buttons are TXT, PDF, DOCX, SRT and VTT. Copying the text to your clipboard works too.

Try it on your own file

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

Getting a cleaner transcript

Speech models are good listeners, but they cannot fix a bad recording. Most of what separates a clean transcript from a messy one happens before you press record.

  • Get close to the microphone. A phone held a hand’s width from someone’s mouth beats a laptop across the table.
  • Choose a quiet, soft room. Carpet, curtains and sofas absorb echo; bare walls and glass bounce it around.
  • Avoid music under speech. Background music is one of the most common causes of garbled lines.
  • Say unusual names and terms clearly once, then check them in the transcript afterwards.
  • Prefer the original file to a copy that was converted several times, which can blur consonants.

Even with a good recording, expect a draft rather than a finished document. Names, numbers, acronyms and unusual words are where the text most often needs a human eye. The Edit view is built for exactly that, and timestamps stay in step with the audio while you fix the words.

What “without uploading” really means

Your recording is opened by your browser and passed to a background worker that runs the speech model on your own processor. The transcript is written to storage inside your browser, so it is still there next time you visit. No request the page makes ever carries your audio or text. The server that hands out the model files only sends files to you. If you would like proof, our guide on whether online transcription is private shows how to check with your browser’s developer tools.

One practical consequence: because nothing is uploaded, there is no upload wait. A large video starts working as soon as you pick it, which is a real time saver on a slow connection.

Long files and slower devices

Running the model on your own device means your device sets the pace. On a computer, Fast handles up to 3 h, Standard handles up to 2 h, High accuracy handles up to 1 h. On phones it's half of that. If a file is longer than your chosen model allows, the workspace says so and suggests a lighter model or splitting the file. Files expected to take more than about 45 min also ask before they start, so you can pick a faster model instead.

A few habits keep things smooth. Plug a laptop or phone in for a long file. Close other heavy tabs while it works. If a lecture or meeting runs past your limit, cut it at a natural break, such as a pause between topics, and transcribe the parts one after another. The text is easy to join afterwards.

No file yet? Record instead

The same workspace can record from your microphone. A live draft appears while you speak (the speech to text page has the details), and when you press Stop the whole recording is transcribed again with your chosen model for the final text. The live draft is a convenience for watching your words appear; the final pass is the one to keep.

Recording is handy for voice memos and quick notes, a quick dictation or a meeting where you have everyone’s permission. Always ask before recording other people. Rules about recording conversations differ between countries, and between states in the US and provinces in Canada, so if you are unsure, get consent first.