Lingo 2.0 🗣️

Local speech-to-text powered by whisper.cpp. No cloud APIs, no costs, complete privacy.

This Lingo 2.0 project contains both a Web App and a GTK app, both of which use Whisper, but are separate apps. The web app is in folder web-app and the GTK app is in the gtk-app folder.

The Web App is almost identical to the original Lingo, under this same Github account (by Clay Ferguson), except the oridinal Lingo uses browser-based Speech API (for Voice Input) rather than Whisper.

For browser-based Speech I do recomment Lingo, rather than Lingo 2.0, just because, if you're already in a browser, there's no reason to use Whisper.

GTK App Screenshot

Web App Screenshot

Projects

This mono-repo contains two applications that provide different ways to use whisper.cpp for voice input:

Project	Description
web-app	Browser-based TTS/STT with a FastAPI backend. Access via http://localhost:8009
gtk-app	Linux desktop app for system-wide voice typing. Speaks into any focused application

Both projects share the same whisper.cpp engine located in whisper-model/.

🔧 First setup Whisper AI

setup-whisper.sh (project root)

One-time setup that:

1. Clones the whisper.cpp repository
2. Builds the whisper-cli binary using cmake
3. Downloads the base.en model (~150MB)

🧠 Upgrading the AI Model

This project uses the base.en model by default, which offers a good balance of speed and accuracy. If you need better accuracy (at the cost of speed) or faster performance (at the cost of accuracy), you can switch to a different model.

Available Models

Model	Size	Speed	Accuracy	Best For
tiny.en	~75MB	Fastest	Decent	Quick testing, low-powered devices
base.en	~150MB	Fast	Good	⭐ Default - good balance
small.en	~500MB	Medium	Better	Improved accuracy without too much slowdown
medium.en	~1.5GB	Slower	Great	High accuracy needs
large	~3GB	Slowest	Best	Maximum accuracy, multilingual

Note: The .en suffix means English-only models, which are smaller and faster. The large model is multilingual (no .en variant).

How to Switch Models

You need to edit two files:

1. setup-whisper.sh (line ~91)

Change the model name in the download command:

# Change from:
./models/download-ggml-model.sh base.en

# To (for example, small.en):
./models/download-ggml-model.sh small.en

2. Set Python Variable

Both the 'gtk-app' and the 'web-app' have this same variable definition which tells it which whisper model to use.

# Change from:
WHISPER_MODEL = WHISPER_DIR / "whisper.cpp" / "models" / "ggml-base.en.bin"

# To (for example, small.en):
WHISPER_MODEL = WHISPER_DIR / "whisper.cpp" / "models" / "ggml-small.en.bin"

3. Re-run whisper setup and restart

# Download the new model
./setup-whisper.sh

Next you can restart the app.

Tip: You can have multiple models downloaded. Just change whisper_server.py to switch between them without re-downloading.

Quick Start

1. Build whisper.cpp and download the model:

   ./setup-whisper.sh

2. Run whichever app you prefer:

   # Web app (browser-based)
   cd web-app && ./run.sh
   
   # GTK app (Linux desktop)
   cd gtk-app && ./run.sh

Requirements

• Linux (Ubuntu/Debian tested) or macOS
• Python 3 with venv support
• ffmpeg
• Build tools (cmake, git, build-essential)

See each project's README for additional dependencies.