Local voice-to-text for Windows.
Hold a hotkey or just start talking; the transcript gets pasted where your cursor is.

---

Features

• Push-to-talk hotkey (default F9) that works in any app
• Voice activation mode (Silero VAD) if you don't want to hold a key
• Pick a specific microphone, or let the OS default route
• Whisper runs locally via whisper.cpp — audio stays on your machine
• CUDA acceleration if you have an NVIDIA GPU, CPU fallback otherwise
• Five Whisper model sizes: tiny, base, small, medium, large-v3-turbo, plus a local Models library to delete/swap them
• Optional DeepL translation into a target language of your choice
• Find & replace rules + an initial prompt passed to Whisper, applied to every transcript
• Hands-free paste — pin a target window and Echo focuses it, pastes, and returns you to whatever you were on. Works even when the target is minimized
• Auto-submit — press Enter after paste, handy for Claude / ChatGPT / Discord
• Rolling transcription history with copy-to-clipboard
• Built-in auto-updater: new releases download in the background and install with one click
• Lives in the system tray; can start at login

Install

Download the latest installer from the Releases page and run it.

On first launch Echo downloads the selected Whisper model (base by default) and, if you have an NVIDIA GPU and pick the CUDA backend, the matching whisper.cpp binary. From v2.2.0 onward, Echo checks for its own updates on launch and installs them from a single Restart & install button in the About tab.

Usage

Push-to-talk (default):

1. Hold F9 (or whatever hotkey you set).
2. Speak.
3. Release. Transcript is pasted into the focused text field.

Voice activation:

In Settings → Hands-free, toggle Voice activation on. Echo then listens continuously and transcribes each utterance as you speak. The mic stays live; the hotkey is disabled while this is on.

Hands-free paste into a specific window:

In Settings → Hands-free, pick a Target window. Every transcript after that goes to that window — Echo focuses it, pastes, and returns focus to whatever you were actually on. Minimized targets are un-minimized for the paste then re-minimized afterward. The pin is keyed by process ID so title changes (Chrome tab switches, Notepad edits) don't break it, and is cleared only when the pinned process actually dies. The pin is intentionally volatile across app restarts — pick it again next session.

Enable Auto-submit alongside it to send Enter after the paste, so voice-driving a chat app is literally hands-free.

Settings

Organised into six tabs in the Settings window (tray → Settings).

General

Setting	Default	Notes
Push-to-talk hotkey	F9	Any key or modifier combo
Exit shortcut	Ctrl+Alt+Q	Global quit
Microphone	System default	Pick a specific input device; falls back to default if unplugged
Start at login	Off

Model

Setting	Default	Notes
Whisper model size	base	tiny / base / small / medium / large-v3-turbo
Language	Auto-detect	Pin one for a small speedup and to enable the translate-skip optimisation
Compute backend	Auto	CPU or CUDA; auto-selected based on hardware
Downloaded models	—	List, delete, and re-download cached models

Post-processing

Setting	Default	Notes
Initial prompt	—	Biases Whisper's style (custom vocabulary, punctuation, tone)
Translate to	Off	Sends the transcript to DeepL for translation
DeepL API key	—	Only needed when a translation target is set. Free keys end in :fx
Find & Replace	—	Case-insensitive rules applied last; supports \n / \t

Hands-free

Setting	Default	Notes
Target window	None	Pin a specific window for all transcripts (by PID, volatile across restarts)
Voice activation	Off	Always-on mic; disables the push-to-talk hotkey
Auto-paste transcript	On	Off = copy to clipboard only
Auto-submit	Off	Press Enter after paste; requires auto-paste

History

Rolling last 50 transcripts, persisted to %APPDATA%\Echo\history.json. Copy each back to the clipboard, delete individually, or clear all. Toggle off to stop saving without wiping existing entries.

About

Current version, platform and GPU badges, Check for updates button, and a "What's new" list of recent releases linking to the GitHub compare view for each version.

Build from source

Requires Node.js 20+ on Windows.

git clone https://github.com/Kenshiin13/echo.git
cd echo/electron
npm install --legacy-peer-deps
npm run dev

Package an installer:

npm run dist:win

Pushing a v* tag runs the release workflow and publishes a GitHub Release with the built artifacts (installer, latest.yml, blockmap — electron-updater needs all three).

Stack

• Electron 33, React 18, Vite 6, Mantine
• whisper.cpp via nodejs-whisper for transcription (model kept resident in a local whisper-server subprocess)
• Silero VAD via @ricky0123/vad-web for voice activation
• DeepL API (optional) for translation
• koffi for GetAsyncKeyState-based hotkey polling and the SetForegroundWindow + AttachThreadInput focus dance that powers the Hands-free paste
• Electron's desktopCapturer + @nut-tree-fork/nut-js for window enumeration and keystroke simulation
• electron-updater for the built-in updater
• electron-store for settings, electron-builder for packaging

License

This project is licensed under the GNU GPL v3.0 - see the LICENSE file for details.