Record or drop in an audio file and get a transcript — generated entirely on your own device by on-browser AI. Nothing is uploaded, and there's no per-minute cost.
⚙ detecting…🔒 no upload💸 no API cost📴 works offline after load
or
Press Record or upload a file to begin. The AI model downloads once on first use.
0 words
How private on-device transcription works
1
Capture audio
Record from your mic or upload a file. It's read straight into memory in your browser.
2
AI runs locally
OpenAI's Whisper model runs on your device's own processor (WebGPU or WebAssembly) — no server.
3
Get your text
The transcript appears for you to edit, copy or download. Nothing was ever uploaded.
Why this matters
Cloud transcription services upload your audio and charge per minute. EchoScribe does neither: the model downloads once (about 40 MB for the tiny model), caches in your browser, and from then on every transcription is free, private and even works offline. It's a clear example of what modern browsers can do — real AI, no cloud bill, no data leaving your device.
Frequently asked questions
Is my audio uploaded anywhere?
No. The Whisper model runs entirely in your browser. Your audio never leaves your device and there's no account. The only network use is a one-time download of the model itself.
How is it free with no API cost?
The AI runs on your own device, not a cloud server, so there's no per-minute fee. The model downloads once (~40 MB), caches, then works offline.
How accurate is it?
It uses OpenAI's Whisper (tiny English by default). It handles clear speech well; accuracy depends on audio quality and accent. A good mic and low background noise help a lot. Switch to the Base model for more accuracy.
Does it work offline?
Yes — after the first load. Once the model has cached, transcription needs no internet at all.
Note: first use downloads the AI model (about 40–145 MB depending on the model you pick), which can take a moment on slower connections. Best performance is on a desktop/laptop with a modern browser; very old devices may be slow or unsupported. Transcripts are AI-generated and may contain errors — review before relying on them.