How LocalTranscribe Works

A technical explanation of how your files stay private.

Browser-Side Processing

LocalTranscribe runs entirely in your web browser. When you load a file, it stays on your device — the AI transcription happens locally using WebAssembly-based models. No audio data, no file content, and no transcripts ever reach a server.

Technical Implementation

These are the specific technologies that make local transcription possible:

Transformers.js

Runs Whisper AI models directly in the browser using JavaScript. Enables state-of-the-art machine learning without a backend server.

Whisper (via ONNX Runtime Web)

OpenAI's Whisper speech recognition model, compiled to WebAssembly and executed in-browser via ONNX Runtime. This is the engine doing the actual transcription.

FFmpeg.wasm

A WebAssembly build of FFmpeg used to extract audio tracks from video files and handle format conversion — all on your device.

File API

The browser's built-in File API reads your file into memory locally. No network request is made to load your content.

SharedArrayBuffer & Web Workers

FFmpeg.wasm and the ML models run in background threads (Web Workers) using SharedArrayBuffer for efficient memory sharing. This keeps the UI responsive during transcription.

What This Means for Privacy

  • Your files are never uploaded to any server
  • No audio content reaches any network endpoint
  • Transcripts are generated and stored only on your device
  • No file names, durations, or metadata are logged or transmitted
  • No third party has access to what you transcribe

How to Verify

You can confirm this yourself:

  1. Open browser developer tools (F12 or right-click → Inspect).
  2. Go to the Network tab.
  3. Load a file and start transcription.
  4. Watch the Network tab — you will see no requests containing your file data.

You can also disconnect from the internet after the page loads. Transcription still works.

Trade-offs

Because processing happens on your device, transcription speed depends on your CPU and available memory. Older or lower-powered devices will take longer than server-based tools. Very large files (over 2 GB) may cause memory pressure on some devices.

Ready to transcribe privately?

Transcribe a file now