Your voice stays yours
Local Transcription
Record up to 30 seconds with your microphone or upload a short clip, and a small open-source speech model turns it into text — entirely inside your browser, in English, German, or auto-detected.
Local Transcription
Local processing
Model not loaded
First use needs internet and about 150 MB. Loading may reuse cached files.
A short excerpt from JFK’s inaugural address. Loading this clip does not download the model.
Preview the example and its words
“And so, my fellow Americans: ask not what your country can do for you — ask what you can do for your country.” Actual model output may differ.
Your audio stays in this browser and is never sent to Best-AI.org or any other server. The model is fetched once from its disclosed source, then inference runs fully offline on your device.
What this experiment can do
Short, clear speech becomes searchable text with word-level segments you can copy. It is a tiny model (39M parameters), not a professional transcription service — accents, music, overlapping voices, and background noise all reduce accuracy.
Model transparency
- Model
- Xenova/whisper-tiny
- Hosting
- Fetched once from Hugging Face (pinned revision), then runs on-device from the browser cache
- Download
- ~150 MB on first use
- Runtime
- WASM in a Web Worker
- License
- Apache-2.0