Your chat. Your device.
The whole conversation.
Including the voice notes.
Turn a WhatsApp export into a readable conversation, with each voice message transcribed in its original place.
01 / Bring your conversation
From voice to text
In WhatsApp, choose Export chat → Include media. Photos and videos are skipped, so large exports are fine. Chat text and voice messages can total up to 512 MB.
Choosing the language is the most reliable. Automatic detection listens to the start of each voice message.
Devices with a capable graphics card use Whisper large-v3-turbo (1.4 GB download, most accurate). Others use Whisper small (240 MB).
Your conversation and audio stay in this tab. The first transcription downloads the Whisper speech model (240 MB or 1.4 GB, kept for next time); no account is needed. Keep this page open while it works.
02 / Keep the conversation
Your readable chat
Voice message details
Review transcripts for accuracy. JSON keeps the original chat, message metadata, transcription segments, and errors for future formatting.
Export with media
Bring the chat text and its voice recordings.
Let your device listen
Whisper turns voice messages into text locally.
Take the whole story
Download the conversation and reusable JSON.
Privacy, built in
Your export never leaves your device.
VoxPad reads your files and transcribes audio in your browser. No audio, chat text, filenames, or transcripts are uploaded to, cached on, or stored on a server. There is no upload endpoint, account, or analytics.
Your conversation stays in this tab’s memory. Reloading or closing the page releases that working data; downloads are saved to your device when you choose. Only the speech model and runtime are cached in your browser for reuse.
GitHub serves the app, and Hugging Face serves the speech model. These downloads expose ordinary connection information, such as your IP address, to those hosts. They do not receive your conversation.
This is enforced, not only promised: the page’s security policy lets your browser connect to this site and Hugging Face only, and the speech model is used only if it matches the exact version this app was tested with.