Realtime Subtitles

Realtime Subtitles is a Python-based script that captures system audio in real-time, transcribes it using OpenAI Whisper, and optionally translates the transcribed text. The subtitles are displayed in an overlay window on your desktop.

Features

• Captures system audio using loopback devices.
• Transcribes audio in real-time using OpenAI Whisper.
• Supports optional translations using Google Translate or DeepL.
• Customizable subtitle overlay with font, color, size, and transparency options.

How It Works

1. Captures system audio using Python libraries (soundcard and sounddevice).
2. Sends audio chunks to an OpenAI Whisper API running in a Docker container.
3. Displays the transcription as subtitles in a transparent overlay window.
4. Optionally translates the transcription using Google Translate or DeepL.

Requirements

• Python 3.8+
• NVIDIA GPU for running Whisper API (optional for GPU acceleration)
• Docker (for hosting the Whisper API)

Installation

1. Clone the repository:

   git clone https://github.com/<your_username>/realtime_subtitles.git
   cd realtime_subtitles

2. Install dependencies:

   pip install -r requirements.txt

3. Set up and run the Whisper API:

   • Install Docker if not already installed.
   • Create and start the Whisper service:

     docker-compose up -d

4. Run the script:

   python main.py

Configuration

The config.json file allows you to customize the following:

• font_name: Font used for subtitles (default: Comic Sans MS).
• font_size: Font size (default: 24).
• font_color: Subtitle text color (default: yellow).
• translation: Enable/disable translation (default: False).
• deepl_api: API key for DeepL translation (if enabled).
• language: Target language for translation (default: english).
• audio_threshold: Minimum audio level to trigger transcription (default: 0.01).
• text_expiry: Time in seconds before the subtitles fade (default: 3.0).
• window_opacity: Transparency of the subtitle window (default: 0.8).

Docker Compose

The docker-compose.yml file is included for setting up the Whisper transcription service with GPU support. It uses the onerahmet/openai-whisper-asr-webservice Docker image.

Dependencies

The following Python libraries are required:

• soundcard
• sounddevice
• numpy
• wave
• requests
• googletrans==4.0.0rc1
• deepl
• langdetect
• tkinter

Contributing

Feel free to fork the repository and make your own changes! Contributions are always welcome.

Disclaimer

This is my shot at this and migth not work perfectly, i invite you to make pull requests and contribute to make this better. (I am kind of a beginner and i was not able to fix that the sometimes it skips audio)

License

MIT License