KeySpeak

The problem with dictation software

Most dictation tools are either baked into an OS (and mediocre), subscription-based cloud services (and privacy nightmares), or enterprise products (and priced accordingly). What's missing is a small, fast, local tool that does one thing: hear what you say and type it where your cursor is.

KeySpeak fills that gap. It runs in the background, triggers on a hotkey, transcribes locally using on-device speech recognition, and outputs directly to whatever input field or document has focus. That’s the entire product.

How it works

KeySpeak sits in the system tray and does nothing until you ask it to. Press your configured hotkey to begin recording. Speak. Press again to stop. Your transcription appears at cursor position in whatever application you were using — a browser text box, a document, a chat input, anything.

Transcription happens on-device using a local speech model. No audio is sent to any server. No account required. Works without an internet connection.

Hotkey Configuration

All hotkeys are remappable. The defaults above are chosen to avoid conflicts with common applications. The tray icon shows recording state with a color indicator so you always know when the mic is live.

Privacy model

Audio is processed entirely on your machine using a bundled local speech model. Nothing is stored, logged, or transmitted. The app requests microphone access once and only activates it during an active recording session.

There is no account, no analytics, no telemetry, and no connection to any remote service during normal operation. Updates are delivered directly — no update agent running in the background.

Design goals

• Launch time under 500ms from hotkey press to recording-ready
• Works in any text input across any application
• Minimal RAM footprint when idle
• No installer — runs as a single portable executable
• No account, no sign-in, no forced updates