Documentation

SpeakText Docs

SpeakText is a local-first dictation app for Windows. Linux and Mac are coming soon. Run it in the tray, press a global keyboard shortcut from any app, and speak to type text directly at your cursor. GPU is recommended for smooth fully offline realtime transcription, and CPU is supported too.

Quick Start

The simplest workflow is tray app running, cursor in the right place, shortcut on, speak, shortcut off.

  1. 1Run SpeakText. It lives in the tray, so it can keep working while you use other apps.
  2. 2Click into any app where you want text to appear, like a browser, editor, chat box, document, or email.
  3. 3Press the global keyboard shortcut. The default shortcut is Control+Shift+Space.
  4. 4Speak normally. SpeakText types your words at the active cursor.
  5. 5Press the same shortcut again to stop listening when you are done.

Set The Shortcut

The shortcut is global, so you can trigger SpeakText while another app is focused. The default is Control+Shift+Space.

Click the shortcut field in SpeakText.
Press the key combination you want to use.
Use Backspace or Delete while focused in the field to clear it.
Pick Toggle if you want the shortcut to start and stop listening from any app. Pick Push to talk if you want SpeakText to listen only while the shortcut is held.

Listening And Stopped

Listening

SpeakText is active. You can be in any app, as long as your cursor is where you want text to land. SpeakText may be waiting for speech, recording your voice, or transcribing what you just said.

Stopped

SpeakText is idle in the tray. The global shortcut can start it again, but no text will be inserted until listening is on.

Performance And Hardware

SpeakText is designed for fully local, offline realtime transcription. A GPU is recommended because it gives the smoothest low-latency experience while your speech is being converted to text on your own machine.

Recommended: GPU

Use a CUDA-capable GPU when possible. This is the best path for responsive realtime dictation, especially with larger models.

Supported: CPU

SpeakText can run on CPU too. It remains local and offline, but transcription may be slower depending on your processor and selected model size.

How Time Saved Is Calculated

SpeakText estimates how long the same words would have taken to type, then subtracts how long you spent dictating them.

estimated typing time = words / 30 WPM

time saved = estimated typing time - dictation time

Example: 120 dictated words at a 30 WPM typing baseline equals 4 minutes of typing. If you dictated those words in 1 minute, SpeakText shows about 3 minutes saved. If dictation takes longer than typing, the value can be negative.

Simple Basics

Model size

Tiny is fastest. Base and Small can improve accuracy but may need more computer power, especially on CPU.

Startup

Turn on Startup if you want SpeakText available automatically when Windows starts.

Privacy

Dictation is designed to run locally. Your spoken text is typed into your active app instead of a cloud editor.

Compare workflowsRead changelog