FasterWhisper Real-Time Transcription

Real-time desktop audio transcription using OpenAI Whisper for Arch Linux.

Features

• Real-time desktop audio capture from any application
• High-accuracy Spanish transcription using Whisper large-v3
• CUDA GPU acceleration for fast processing
• Automatic clipboard integration
• Word-by-word progressive display
• WebSocket server-client architecture

Quick Start

1. Clone the repository:

git clone https://github.com/yourusername/FasterWhisper.git
cd FasterWhisper

2. Run the setup script:

chmod +x setup_arch.sh
./setup_arch.sh

3. Start transcription:

trs

System Requirements

• OS: Arch Linux
• GPU: NVIDIA GPU with CUDA support (recommended)
• Audio: PulseAudio or PipeWire
• RAM: 8GB minimum (for large-v3 model)

Dependencies

The setup script will install:

• CUDA toolkit and NVIDIA drivers
• PipeWire/PulseAudio support
• Python conda environment with all required packages
• System utilities (clipboard, notifications)

Configuration

Copy .env.example to .env and configure:

cp .env.example .env
# Edit .env with your OpenRouter API key (optional for LLM refinement)

Main configuration in config.json:

• Audio source selection
• Whisper model settings
• Language preferences
• Processing intervals

Usage

# Start transcription (auto-starts server)
trs

# Get help
trs --help

# Manual server start
conda activate trs
python whisper_server.py

# Manual client start (separate terminal)
python trs_client.py

Troubleshooting

Audio issues:

# List audio sources
pactl list sources

# Test audio capture
trs --test

CUDA issues:

# Check NVIDIA driver
nvidia-smi

# Test CUDA availability
python -c "import torch; print(torch.cuda.is_available())"

Environment issues:

# Recreate conda environment
conda env remove -n trs
conda env create -f environment.yml

Architecture

• whisper_server.py - Main transcription server with WebSocket
• trs_client.py - Display client with clipboard integration
• audio_capture.py - PulseAudio desktop audio capture
• utils.py - Configuration and utilities
• trs - Global command wrapper script

License

MIT License - See LICENSE file for details.