TalkyTexty

Voice to text, anywhere on your desktop. Speak and your words appear wherever your cursor is.

Runs entirely offline using local Whisper models — no API keys, no cloud, no data leaves your machine.

---

How it works

1. Press a global hotkey from any application
2. Speak into your microphone
3. Press the hotkey again (or release it in push-to-talk mode)
4. Your transcribed text is typed into the active window

TalkyTexty runs in your system tray and works with any application — text editors, browsers, chat apps, terminals, IDEs.

Features

• Global hotkeys — start/stop recording from any application
• Toggle or push-to-talk — hold to record or press to toggle
• Local transcription — whisper.cpp models run entirely on-device
• Text injection — transcribed text is typed into the active window automatically
• Model management — download and switch between Whisper models of different sizes
• Visual feedback — floating overlay shows recording status with animated soundwave
• System tray — runs in the background, accessible from the menu bar
• Customizable — hotkeys, overlay style, recording mode, and more

Install

Prerequisites

Requirement	Version	Notes
Rust	1.80+	rustup update stable
Node.js	20+
pnpm	9+	npm install -g pnpm
cmake	any	Required for whisper.cpp compilation

macOS: xcode-select --install && brew install cmake

Windows: Visual Studio Build Tools with C++ workload (cmake included)

Linux: sudo apt install libasound2-dev libgtk-3-dev libwebkit2gtk-4.1-dev libayatana-appindicator3-dev cmake

Build from source

git clone https://github.com/codewithtim/talkytexty.git
cd talkytexty
pnpm install
pnpm tauri build

The first build takes several minutes (compiles whisper.cpp and all Rust dependencies).

The built app is in src-tauri/target/release/bundle/:

• macOS: macos/TalkyTexty.app — drag to your Applications folder
• Windows: nsis/TalkyTexty_x.x.x_x64-setup.exe — run the installer
• Linux: appimage/TalkyTexty_x.x.x_amd64.AppImage or deb/ — install as usual

First launch

1. Open TalkyTexty — it starts in the system tray (menu bar on macOS)
2. Click the tray icon and select Show Settings
3. Grant permissions when prompted:
   • macOS: Microphone access and Accessibility (System Settings > Privacy & Security)
   • Windows: No special permissions needed
   • Linux: Ensure ALSA is working (arecord -l lists devices)
4. Navigate to Models and download a model (Small English Q5 recommended)
5. Open any text editor, press Cmd+Shift+Space (macOS) or Ctrl+Shift+Space, and speak
6. Press the hotkey again to stop — your text appears in the editor

Choosing a model

Model	Size	Speed	Accuracy	Best for
Base English (Q5)	~60 MB	Fastest	Good	Quick testing
Small English (Q5)	~190 MB	Fast	Better	Daily use (recommended)
Small English	~466 MB	Moderate	Better	Best quality for size
Large V3 Turbo (Q5)	~547 MB	Slower	Best	Maximum accuracy, multilingual

Default hotkeys

Action	macOS	Windows / Linux
Toggle recording	Cmd+Shift+Space	Ctrl+Shift+Space
Push-to-talk	Cmd+Shift+V	Ctrl+Shift+V
Open target selector	Cmd+Shift+T	Ctrl+Shift+T
Open settings	Cmd+Shift+,	Ctrl+Shift+,

All hotkeys are customizable in Settings. Push-to-talk is disabled by default.

Recording modes

Toggle (default): Press the hotkey to start recording. Press again to stop and transcribe.

Push-to-talk: Hold the hotkey to record. Release to stop and transcribe. Enable in Settings > Recording.

Text injection

Two methods for inserting text (configurable in Settings):

• Simulated Keystrokes (default) — types text character by character. Works in most applications.
• Clipboard Paste — pastes via Cmd+V / Ctrl+V. Faster for long text. Restores your previous clipboard content.

---

Development

Built with Tauri v2 (Rust backend + React frontend).

Running in dev mode

pnpm tauri dev

Starts the Vite dev server and launches the app with hot reload. Subsequent builds are incremental and fast.

Testing

# Rust tests
cd src-tauri && cargo test

# Frontend tests
pnpm test

# All quality gates
cd src-tauri && cargo test && cargo clippy -- -D warnings && cd .. && \
  npx tsc --noEmit && pnpm lint && pnpm format:check

Project structure

src/                           # React frontend (TypeScript)
  pages/                       #   Route pages (settings, overlay, picker)
  components/                  #   UI components (soundwave, hotkey recorder, etc.)
  hooks/                       #   React hooks (recording, models, preferences)
  types/                       #   Shared TypeScript interfaces
src-tauri/                     # Rust backend
  src/
    lib.rs                     #   App entry, tray, hotkey handler
    audio/                     #   Audio capture (cpal) and resampling (rubato)
    transcription/             #   Whisper engine and model registry
    injection/                 #   Text injection (keyboard, clipboard)
    preferences/               #   User preferences persistence
    commands/                  #   Tauri IPC command handlers
  tests/                       #   Rust test suites

Troubleshooting

"No input device available" — Check microphone permissions. macOS: System Settings > Privacy & Security > Microphone.

"Failed to load whisper model" — Delete the model in Settings > Models and re-download.

Hotkeys not working — macOS requires Accessibility permission. System Settings > Privacy & Security > Accessibility > add TalkyTexty.

Text not appearing — Some apps block simulated keystrokes. Try Clipboard Paste mode in Settings.

Disclaimer

This project is entirely vibe-coded using Claude and is not a representation of my coding abilities.

License

MIT — see LICENSE for details.