toolcrate
All tools →
🧠 Real AI · runs on your device

Speech to Text Transcriber

Transcribes a recording or an audio file using OpenAI's Whisper model, running entirely on-device (~78MB one-time download, then works offline). No audio is ever uploaded anywhere.

Loading Whisper speech model (~78MB, one time)…

How it works

Transcription runs entirely on-device using OpenAI's Whisper model (the tiny English variant) loaded through Hugging Face's Transformers.js library — a one-time ~78MB download that's then cached in the browser. Recorded or uploaded audio is decoded into a mono 16kHz signal (resampled automatically by constructing the AudioContext at that exact rate, since that's what Whisper expects) before being fed to the model.

FAQ

Does this work with languages other than English?
No — it uses whisper-tiny.en, an English-only Whisper model chosen because it's the smallest and fastest to download; audio in another language will transcribe poorly or not at all.
Why does the first transcription take so much longer than later ones?
The ~78MB Whisper model has to download and load into your browser first; once it's cached, every transcription afterward in that session starts almost immediately.

Free, unlimited, no account. If this saved you a paid tool, a small thanks is appreciated.

Say thanks — $3