Transcribe audio in your browser, privately
Updated
Short answer: As of November 2026, you can transcribe audio without uploading it by running OpenAI’s open-source Whisper model inside your browser. Voicely does this with transformers.js and WebAssembly: you download the model once (≈ 41 MB or ≈ 77 MB), it is cached, and every recording or file is transcribed on your own device. It works offline afterwards. The trade-off is speed: on older phones it takes roughly 1–3 times the length of the audio.
How it works
Whisper is a speech recognition model released by OpenAI as open source. Small versions of it (whisper-tiny and whisper-base) can run in a web page using WebAssembly. When you choose on-device Whisper in Voicely, the browser downloads the model files once, stores them in its cache, and runs them on your phone’s or computer’s processor.
Your audio is decoded and transcribed inside the page. It is not sent to Voicely or anyone else. The note rules that write the summary and tasks also run in the page.
Small or Better?
“Small” is whisper-tiny, about 41 MB: the default, fastest, good for clear English. “Better” is whisper-base, about 77 MB: slower, more accurate, and the one to pick for Urdu. On both, Urdu and Punjabi accuracy is fair rather than great; that is the honest limit of small models.
Limits to know
It needs a reasonably modern browser and some free memory; very old phones may struggle. Long files take a while. The first use needs internet for the download; after that it works offline. Recordings are kept on your device (you can switch that off), and nothing is backed up to the cloud unless you choose to.
Voicely is free to use in your browser at voicely.danixsoft.com/app, with no account needed. Cloud transcription (the most accurate engine for Urdu) is not connected yet; the app says so.
Comparison
| On-device Whisper | Live captions | Cloud Whisper | |
|---|---|---|---|
| Status | Available | Available (Chrome, Edge, Safari) | Not connected yet |
| Where audio goes | Nowhere: stays in your browser | Your browser’s speech service (Google or Apple) | Groq, when connected |
| Speed | About 1–3× the note’s length on older phones | Real time | Fast |
| Languages | Auto, English, Urdu, Hindi, Arabic, Punjabi | English, Urdu (ur-PK), Hindi, Arabic | Best Urdu accuracy |
| Download | ≈ 41 MB (Small) or ≈ 77 MB (Better), once | None | None |
| Works on | Recordings, uploads, shared notes | Live recording only | Recordings and uploads; needs an account |
Voicely’s transcription engines (November 2026). Other apps change often; check your own.
براؤزر میں نجی ٹرانسکرپشن
On-device Whisper، OpenAI کا اوپن سورس ماڈل، آپ کے براؤزر کے اندر چلتا ہے۔ ماڈل ایک بار ڈاؤن لوڈ ہوتا ہے (تقریباً 41 یا 77 MB)، پھر ہر ریکارڈنگ آپ کے اپنے فون پر تحریر بنتی ہے۔ آواز کہیں اپ لوڈ نہیں ہوتی، اور بعد میں انٹرنیٹ کے بغیر بھی چلتا ہے۔ پرانے فون پر یہ کچھ سست ہے۔
Voicely اردو میںQuestions
- Is on-device transcription really private?
- Yes. The model runs in your browser and the audio is not uploaded. Voicely’s cookieless page analytics never include the content of notes.
- Why is it slower than cloud transcription?
- It runs on your device’s processor instead of a data-centre GPU. On older phones expect roughly 1–3 times the audio length.
- Which languages does it support?
- Auto-detect, English, Urdu, Hindi, Arabic and Punjabi in Voicely’s picker.
Try it on your own voice note
Free, in your browser, no account needed. Record, upload or share a voice note and get text, a summary and tasks.