Audio Transcriber — Voice Notes to Text

Drop a WhatsApp voice note, Opus file, MP3, or WAV and get a text transcript. Everything runs on your device — your audio is never uploaded to a server.

100% Free Nothing Uploaded Works on Mobile

Drop an audio file here or click to select

Opus, MP3, WAV, M4A/AAC, OGG — decoded locally, nothing uploaded

How It Works

This tool replaces a common CLI workflow — converting audio with ffmpeg, then running OpenAI Whisper — with a fully in-browser pipeline:

  1. Decode — your file is read with the Web Audio API (Opus, MP3, WAV, and more).
  2. Prepare — audio is mixed to mono and resampled to 16 kHz for speech recognition.
  3. Transcribe — an English Whisper model runs locally via Transformers.js. Long files are processed in 30-second chunks.

Your audio bytes never leave your device. Model weights are loaded from media.rus.io; transcription still runs entirely in your browser (no Hugging Face or cloud STT API).

WhatsApp Voice Notes

WhatsApp exports voice messages as .opus files. Save the attachment from the chat, drop it here, and read the transcript instead of playing audio — useful when you are in meetings, on calls, or wearing headphones for something else.

Choosing a Model

Fast (tiny) is the default and works well for short, clear English voice messages. Accurate (base) is slower but handles noisy or mumbled speech better. Both are English-only in this version.

Supported Formats

Opus, MP3, WAV, M4A/AAC, and OGG in Chrome, Edge, Firefox, and Safari. If decoding fails, try exporting the file to WAV or MP3 first.

Privacy

No accounts, no server-side transcription API, no storage of your files. Processing happens in memory in your browser tab.

Frequently Asked Questions

Does this upload my audio?
No. Decoding and transcription run entirely on your device. Model weights are fetched from media.rus.io; your audio is never sent to a transcription API.
Why is the first run slow?
The browser still has to load the Whisper model files into memory the first time you transcribe (~40 MB for Fast, ~75 MB for Accurate). Your browser caches them for later visits.
Can it transcribe non-English audio?
Not in v1 — both models are English-only (.en). Multilingual support may come in a future update.
How long can my file be?
There is no hard limit. Audio longer than 30 seconds is split into chunks automatically. Very long files will take proportionally longer on slower devices.