Text & AI Writing
Voice note to text
Record a voice note or open an audio file and get a transcript, subtitles and quick notes. Speech recognition runs inside your browser, so private audio stays on your device.
- Free, no sign-up
- Runs in your browser
- Nothing uploaded
- Runs on your device
1. Record or upload audio
2. Choose the model and transcribe
3. Your text and notes
How it works
Speech recognition here uses Whisper, an open model from OpenAI released under the MIT licence, converted to run in the browser with transformers.js (Apache-2.0) and ONNX Runtime Web (MIT). The code is hosted on beginthings.com. The model files come from Hugging Face the first time you use them and are then cached by your browser, so later runs work without downloading again.
Your audio is decoded and processed on your device and is never sent to a server. That makes it suitable for private notes, interviews and meetings. Accuracy depends on the model and the recording: the English models do well on clear speech, the multilingual tiny model is rougher, especially for Indian languages and accents. Always read the transcript before you rely on it.
The quick notes are made with plain rules, not an AI summary: each sentence becomes a bullet, sentences that sound like tasks are listed as possible action items, and dates, times and amounts are collected. Clips can be up to 10 minutes. Transcribing takes roughly as long as the clip on a typical laptop and longer on an older phone.
Frequently asked questions
Is my audio uploaded?
No. The model runs on your device. The only download is the model itself, once.
Does it work offline?
After the first model download, yes, in most browsers.
Which languages work?
English is best. The multilingual model also handles Hindi and other languages, but accuracy is lower, so check the text.
Why is it slow on my phone?
The model runs on your processor. Use the fast English model, keep clips short and keep the screen on.