Can you transcribe audio offline without uploading it?

Yes. You can transcribe audio offline by running a speech recognition model on your own computer. The audio never has to leave the machine, provided the tool does its work locally. Most offline tools need a model download during setup, so do that while you are online. After that, you can disconnect and transcribe. The guide to transcribing audio to text covers the rest of the process, and the transcription overview links the other guides.

Whisper-based apps run on your computer once they are installed

Several free tools run Whisper locally. Handy describes itself as a free, open source speech-to-text application, and its README says transcription happens on your own computer without sending any information to the cloud ¹. Buzz describes itself as transcribing and translating audio offline on a personal computer ².

Developers who want the same thing in their own software can use whisper.cpp, a C and C++ port of Whisper. Its README includes a video of the model running on an iPhone 13 "fully offline, on-device" ³. That demonstration shows the port can run without a connection. It does not tell you how fast the model runs on your computer.

Check the tool's own description before you trust it. Some apps offer both a local model and a cloud model, and only the local one keeps the audio on your computer.

Setup usually downloads a model, so do it while you are online

Offline tools need a model file on disk, and most download it the first time you set them up. The HEX README, for a Mac dictation app, says you download a local speech model during setup, and that ordinary dictation does not need a cloud transcription service ⁴. Check the setup screen of any offline tool to see what it downloads and when.

Your computer's memory decides which model you can run offline

The model size sets the memory you need. The largest Whisper model needs about 10 GB of video memory ⁵. A computer without that much memory should choose a smaller model, which runs on modest hardware but gives less accurate text. The guide to running Whisper locally gives the setup steps.

Whisper's own README also requires the ffmpeg command-line tool to be installed on your system ⁶. Install it during setup, while you are online. The question on audio formats explains which formats each tool accepts.

Chrome's Live Caption works offline, but test whether it saves text

Google says Live Caption in Chrome works offline, including for audio and video files saved on the computer ⁷. Live Caption shows captions on screen as the audio plays. The fact does not say whether the captions can be saved as text, so test that before you plan a transcript around it. For a finished text file, a Whisper-based app is the more reliable route.

Cloud tools send the audio to a provider, so read their data terms

A cloud service is not offline, and it may keep what you send for a period. OpenAI says that the abuse monitoring logs for its API are kept for up to 30 days by default ⁸. That is a provider policy, and it can change, so read the terms for the plan you use. If the recording is confidential, a local tool avoids the question altogether.

How to confirm a tool works offline before you use it for a real recording

1. Read the description. Look for words such as "offline" or "on your device" in the app's own description, and note which model it downloads.
2. Set up while online. Install the app, let it download its model, and note where the model is stored.
3. Turn off Wi-Fi and Ethernet. Disconnect the computer completely.
4. Transcribe a short test clip. Use a recording you know well, and confirm that the text appears with the connection off.
5. Proofread the output. Check it against the audio, especially names and numbers, before you use the transcript for anything important.

If the test fails, the tool is not working offline on your computer, and the audio may be going elsewhere. Stop and read its settings before you use a real recording.

For a recording that is hard to hear, the offline draft still needs the same checks. The question on low-quality audio explains how to improve the source first. For free routes that use the internet, see how to transcribe audio for free. The word error rate explains how to read an accuracy figure, and the automatic speech recognition entry describes what the tools do.