🎬 Local-LLM-Powered-Subtitle-Translator – SRT Translator

Sublingo is a robust, local LLM-powered tool designed to translate subtitle files (.srt) while maintaining strict formatting. By leveraging Ollama, it ensures your data stays private, secure, and runs entirely on your local machine.

Key Features

• Privacy-Centric: All translations happen locally via Ollama. No data is sent to the cloud.
• SRT-Safe Parsing: Specifically engineered to handle SRT structures, ensuring timestamps and sequence numbers remain untouched.
• Responsive UI: Features a real-time progress bar and status updates, so you’re never left wondering if the app is frozen.
• Customizable Settings: * Choose between different local models (Aya, Gemma, Llama, etc.).
  • Multiple target languages including Persian, French, Russian, and more.
  • Adjustable timeout settings for older hardware.
• Green-Light Workflow: A clear, color-coded interface for easy navigation.

---

Prerequisites

1. Ollama Installed: Download Ollama here.
2. Models Downloaded: Open your terminal and pull the models you want to use:

   ollama pull aya-expanse:8b
   ollama pull gemma3:1b

3. Python Environment: Python 3.9 or higher is recommended.

---

Installation & Setup

1. Clone the repository:

   git clone [https://github.com/developerashkan/Local-LLM-Powered-Subtitle-Translator.git]
   cd Local-LLM-Powered-Subtitle-Translator

How to Use:

Launch Ollama: Ensure the Ollama server is running in the background.

Upload File: Drag and drop your .srt file into the upload box.

Configure Sidebar:

Select your Model (use smaller models like gemma3:1b for faster results on CPUs).

Choose your Target Language.

Translate: Click the green Translate Subtitles button.

Download: Once the progress bar hits 100%, the ⬇️ Download Translated SRT button will appear.

Technical Details

The Translation Prompt Sublingo uses a specialized "Professional Translator" prompt to ensure the LLM doesn't hallucinate or add conversational filler:

"Output ONLY the translated text. Preserve line breaks. Keep it natural and subtitle-friendly."

Progress Handling The app uses a min((i + 1) / total_blocks, 1.0) logic to ensure the Streamlit progress bar never exceeds 100%, even with floating-point math variances, providing a smooth user experience.

Contributing

Feel free to fork this project, submit PRs, or report issues. Suggestions for batch-processing features or additional language support are always welcome!

License: MIT

Author: Ashkan Bahmani