Local STT Speech recognition, running locally.

How this listening desk works

Talk, and watch it land

Press Start listening and speak naturally. The first words usually appear within a second or two, and keep updating while you talk. Press stop when you are done and the engine makes one clean final pass over the whole take.

A take lasts up to 75 seconds

One take holds a little over a minute of audio — enough for a paragraph, a voice note, or a short interview answer. The timer turns amber near the limit and the recording stops itself at 75 seconds. Start a new take whenever you like.

Files are welcome too

Transcribe a file accepts WAV, MP3, M4A, and anything else your browser can decode. The audio is decoded and resampled locally in this tab. The first 75 seconds are transcribed.

Edit the result

The final transcript is an ordinary editable text box. Fix a name, add punctuation, then copy it or save it as a .txt file. The take itself can be replayed and downloaded as a WAV.

Privacy, please

Microphone audio never leaves this tab. There is no upload step and no account. The model downloads once and is stored in your browser cache, so later visits can transcribe without a network connection.

For the cleanest words

Stay within a hand-span of the microphone, keep the room reasonably quiet, and speak at a normal pace. The model is English-only and most comfortable with clear, continuous speech rather than shouted single words.

Voice in. Text out.

Listening engine

Not loaded

Checking for a cached model…

LOCAL STT // STREAMING TRANSCRIPTION DESK
Microphone not connected

Ready when you are.

Load the model, then press start.

INPUT −∞ dB
TAKE 0:00.0 75s cap
TAKE SCOPE

Nothing heard yet.

Your words will land here, one little block at a time.