Audio to Text

Transcribe speech from audio or video using an on-device AI model — nothing leaves your browser

Drop an audio or video file here, or click to browse
MP3, WAV, M4A, MP4, WebM — shorter clips transcribe fastest
Powered by Whisper running fully client-side via Transformers.js. Accuracy depends on audio clarity and background noise.

About this tool

This transcribes speech using OpenAI's Whisper model, loaded and run entirely in your browser through Transformers.js — audio is decoded to mono 16kHz PCM on-device and fed straight into the model, with no server ever involved. You can pick a smaller English-only model for speed or a multilingual one that also supports translating the source language into English.

Which model size should I pick?

"Tiny" is fastest and smallest to download but makes more mistakes on unclear audio; "Base" is slower to load and run but noticeably more accurate. If your audio isn't in English, use the multilingual Tiny model, which also unlocks the translate-to-English option.

Do I need an internet connection for every transcription?

Only the first time you use a given model size — after it downloads (30–150MB depending on size) it's cached by your browser, so later transcriptions with that same model run without re-downloading anything.

Is my audio or video file ever uploaded anywhere?

No. Decoding and transcription both happen locally using the Web Audio API and an in-browser AI model — the file itself never leaves your device.