MVP-Echo Toolbar

Industry-grade voice-to-text for Windows 11. Record, transcribe, clipboard — in under a second. No accounts, no cloud, no subscriptions.

MVP-Echo sits in your system tray and turns speech into text with a single keyboard shortcut. GPU transcription hits 99% accuracy in under 300ms. Local CPU mode works offline with zero setup. Your audio never leaves your machine.

---

Quick Start

Download and run. No installer, no setup, no account required.

1. Grab MVP-Echo Toolbar 3.0.3.exe from Releases
2. Run it — it's a portable executable
3. Press Ctrl+Alt+Z to start recording
4. Talk, then press Ctrl+Alt+Z again to stop
5. Wait for the ding — your text is already in the clipboard
6. Ctrl+V to paste anywhere

That's it. The app is invisible during normal use. It lives in your notification area — the icon changes color as it records and processes. Click it to see your last transcription or adjust settings.

Why MVP-Echo

• Fast — GPU transcription returns in under 300ms. Local CPU under 2 seconds.
• Accurate — Powered by NVIDIA Parakeet TDT models. 99% accuracy on English speech.
• Private — Audio stays on your machine or your LAN. No telemetry. No cloud. No data collection.
• Simple — One keyboard shortcut. One portable exe. No Python, no dependencies, no configuration.
• Free and open source — Apache 2.0 licensed. Use it, modify it, share it — just give credit.

Features

• GPU Transcription — 99% accuracy, under 300ms via a hosted GPU server on your LAN
• Local CPU Mode — Works offline with no internet and no setup required
• Model Switching — Choose between English (GPU), Multilingual (GPU), or English CPU
• Auto-Clipboard — Transcriptions are copied automatically with an audio confirmation
• Privacy Mode — One-click toggle to disable the microphone entirely
• System Tray — Minimal footprint, always accessible, never in the way

Tray Icon States

Color	Meaning
Blue	Ready
Red	Recording
Yellow	Processing
Green	Copied to clipboard

Models

Model	Engine	Accuracy	Speed	Internet Required
English	GPU (Parakeet TDT 0.6b v2)	99%	<300ms	LAN only
Multilingual	GPU (Parakeet TDT 0.6b v3)	97%	<500ms	LAN only
English CPU	Local (Parakeet TDT via sherpa-onnx)	80%	<2s	No

Architecture

Windows 11                             LAN Server (Docker)
+---------------------------+          +----------------------------+
|  MVP-Echo Toolbar         |          |  mvp-stt-docker            |
|  (system tray app)        |  HTTP    |  (GPU transcription)       |
|                           | -------> |                            |
|  Ctrl+Alt+Z = record      |          |  NVIDIA GPU + CUDA         |
|  ding = clipboard          | <------- |  Parakeet TDT models       |
|                           |  text    |  Port 20300                |
|  Local CPU fallback:      |          +----------------------------+
|  sherpa-onnx (offline)    |
+---------------------------+

MVP-Echo works in two modes:

• Local CPU — Uses sherpa-onnx bundled directly in the app. No server, no internet, no setup. Good for quick notes and offline use.
• Hosted GPU — Connects to a GPU transcription server running on your local network. Deploy the included Docker stack on any machine with an NVIDIA GPU for the best accuracy and speed available.

Setting Up GPU Transcription

The portable exe works immediately with local CPU. For GPU quality:

1. Deploy the STT server on any machine with an NVIDIA GPU:

   cd mvp-stt-docker
   docker compose up -d

   See mvp-stt-docker/ for full setup instructions.

2. Run MVP-Echo Toolbar on your Windows machine

3. Click the tray icon > Settings > set your server endpoint:

   http://<server-ip>:20300/v1/audio/transcriptions
   

4. Enter your API key and click Test Connection

5. Select the English GPU model — it persists across restarts

Build From Source

cd mvp-echo-toolbar
npm install
npm run dev          # Development mode (Vite + Electron)
npm run dist         # Build portable exe

Output: dist/MVP-Echo Toolbar 3.0.3.exe

Note: Building requires binary assets (sherpa-onnx-bin/, sherpa_onnx_models/, ffmpeg.exe) that are too large for git. See Releases for the pre-built executable, or contact us for build asset instructions.

Tech Stack

• Desktop: Electron 28 + React 18 + TypeScript + Vite
• STT Engines: NVIDIA Parakeet TDT (GPU), sherpa-onnx (CPU)
• Styling: Tailwind CSS
• Packaging: electron-builder (portable exe)

Project Structure

mvp-echo-toolbar/       # Electron app (React + Vite + TypeScript)
  app/main/             # Electron main process
  app/renderer/         # React UI
  app/stt/              # Speech-to-text engine adapters
  app/audio/            # Audio capture utilities
mvp-stt-docker/         # GPU server deployment (Docker Compose)

Contributing

Contributions are welcome. Fork the repo, make your changes on a branch, and open a pull request.

If you find a bug or have a feature request, open an issue.

If you build something on top of MVP-Echo, we'd genuinely love to hear about it. We're building more tools in this space and your use case might shape what comes next. Start a thread in Discussions or reach out at mvp-scale.com.

License

Apache License 2.0 -- Copyright 2026 MVP-Scale.com

Use it, modify it, distribute it. Just keep the attribution and let people know what you changed. See NOTICE for details.

MVP-Echo is built on the shoulders of great open-source projects. See THIRD-PARTY-NOTICES for full attribution.