COW-TO-TEXT LIVE TRANSLATOR 🎤

A real-time audio transcription and translation tool that captures live audio from your system, transcribes it using OpenAI's Whisper model, and translates it to your target language.

Features

• 🎤 Real-time audio capture from system audio
• 🗣️ Speech-to-text transcription using Faster Whisper
• 🌍 Multi-language translation using Argos Translate
• ⚡ Multi-threaded pipeline for efficient processing
• 🎨 Colored console output for easy reading
• 🔧 Interactive configuration for model size, chunk size, and language selection

Requirements

• Python 3.11+
• PipeWire/PulsAudio (for audio capture)
• FFmpeg
• CUDA-capable GPU (optional, but recommended for better performance)

Installation

Option 1: AUR (Arch Linux Users)

yay -S cowtotext
# or
paru -S cowtotext

Option 2: Manual Installation from Source

git clone https://github.com/MeIsGaming/cow-to-text.git
cd cow-to-text
python3.11 -m venv venv_fresh
source venv_fresh/bin/activate
pip install -r requirements.txt

Option 3: Quick Setup (with pip directly)

git clone https://github.com/MeIsGaming/cow-to-text.git
cd cow-to-text
pip install -r requirements.txt
python3.11 cowtotext.py

System Dependencies

Before installation, ensure you have:

# Arch Linux
sudo pacman -S python ffmpeg libpulse

# Ubuntu/Debian
sudo apt install python3.11 ffmpeg libpulse0

# Fedora
sudo dnf install python3.11 ffmpeg pulseaudio-libs

Usage

Run the translator:

python3.11 cowtotext.py

The application will prompt you to:

1. Select the Whisper model size (tiny, base, small, medium, large)
2. Choose the audio chunk size (100ms - 2000ms)
3. Select source language
4. Select target language

Once configured, the translator will:

• Capture live audio from your system
• Display transcribed text in the source language
• Display translated text in the target language

Exit: Press Ctrl+C to stop the translator.

Configuration Options

Model Sizes

• tiny: Fastest, lower quality
• base: Balanced
• small: Good quality (works best with pl->en i found out)
• medium: High quality
• large: Highest quality, slowest

Supported Languages

• Afrikaans, Arabic, German, English, Spanish, French
• Italian, Japanese, Korean, Polish, Portuguese, Russian, Chinese

Architecture

The application uses a multi-threaded pipeline:

Audio Capture → Transcription Worker → Translation Worker → Output
                (FFmpeg/PulseAudio)   (Faster Whisper)      (Argos)

• Audio Queue: Raw audio chunks from the system
• Transcribed Queue: Transcribed text from Whisper
• Output Queue: Final translated text

Development

To modify or extend the application:

• Edit language support in the LANGUAGES dictionary
• Adjust threading pool sizes in worker thread creation
• Customize chunk size and overlap for performance tuning

License

MIT License

Contributing

Contributions are welcome! Please create a pull request with your improvements.