livecaption 1.0.0

Real-time audio transcription for video streaming with Firefox browser integration

🎙️ LiveCaption

Real-time audio transcription for video streaming with Firefox browser integration

LiveCaption captures system audio and transcribes it in real-time using state-of-the-art Whisper models. Perfect for Japanese anime, streaming content, and multilingual videos.

✨ Features

• Real-time transcription - See subtitles as you watch
• Firefox integration - One-click recording from browser toolbar
• Multiple Whisper models:
  • Kotoba Whisper v2.0 - Best for Japanese (recommended)
  • Whisper Large v3 - Best for English and 99+ other languages
  • Anime Whisper - Specialized for anime/games
• SRT output - Standard subtitle format for video players
• Voice Activity Detection - Accurate timestamp alignment
• Command-line & browser modes - Use from terminal or Firefox extension

📋 System Requirements

• Operating System: Linux (tested on Fedora, Ubuntu)
• Browser: Firefox 91+ (for browser extension)
• Audio System: PipeWire or PulseAudio
• Python: 3.9 or higher
• RAM: 8GB minimum, 16GB recommended
• GPU: NVIDIA GPU recommended (~4GB VRAM), CPU mode available

🚀 Installation

Method 1: pip (Recommended)

pip install livecaption
livecaption-setup

Note: After pip install, you must run livecaption-setup to register the Firefox native messaging host.

Method 2: From Source

git clone https://github.com/b-tok/LiveCaption.git
cd LiveCaption
./install.sh

First-Time Model Download

• AI models are 1-6GB each and download on first use
• First download takes 5-30 minutes depending on internet speed
• Models are cached locally for subsequent runs

📖 Usage

Browser Extension

1. Install Firefox Extension:

   • Firefox Add-ons Store (recommended)
   • Or download .xpi from GitHub Releases

2. Click the LiveCaption icon in Firefox toolbar

3. Select your settings:

   • Model: kotoba-v2.0 for Japanese, large-v3 for English
   • Audio Source: Usually auto-detected
   • Output File: Where to save the SRT (default: ~/Documents/LiveCaption/recording_<timestamp>.srt)

4. Click "Start Recording" and play your video

5. Click "Stop Recording" to save the SRT file

Command Line

# Basic usage (Japanese content)
livecaption --model kotoba-v2.0 --output subtitles.srt

# English/multilingual content
livecaption --model large-v3 --output subtitles.srt

# Anime/games (Japanese)
livecaption --model anime-whisper --output anime.srt

# List all available models
livecaption --list-models

# Get help
livecaption --help

Workflow:

1. Run the command
2. Start playing audio (YouTube, Netflix, local video, etc.)
3. Press Ctrl+C to stop recording
4. Find your subtitles in the output file

⚙️ Models

Model	Best For	Size	Languages	Recommended Use
kotoba-v2.0	Japanese	~4GB	Japanese	Best for Japanese content
large-v3	Multilingual	~6GB	99+ languages	Best for English/other languages
anime-whisper	Anime/Games	~4GB	Japanese	Anime, visual novels, games
kotoba-v1.0	Japanese	~4GB	Japanese	Older, more stable
medium	Fast	~3GB	Multilingual	Faster but less accurate

Recommendation: Use kotoba-v2.0 for Japanese, large-v3 for everything else.

🗑️ Uninstallation

# Complete uninstall (recommended)
livecaption-uninstall

# Alternative method
python -m livecaption.uninstaller

Note: Manually remove the Firefox extension from about:addons if installed.

📝 Configuration

Settings are stored in ~/.config/livecaption/config.json:

{
  "language": "ja",
  "model": "kotoba-v2.0",
  "device": "auto",
  "output_dir": "~/Documents/LiveCaption",
  "chunk_duration": 30.0
}

📄 License

MIT License - see LICENSE file for details.

🙏 Acknowledgments

• Kotoba-Whisper - Japanese-optimized Whisper
• Anime-Whisper - Anime-specialized model
• OpenAI Whisper - Base transcription model
• Silero VAD - Voice activity detection