📝 Greek SRT Corrector

✨ Automate Greek Subtitle Correction with AI!

A Streamlit-based web application for fast and accurate Greek subtitle correction using a powerful, locally hosted LLM from Ollama.

---

🚀 Key Features

• Greek-focused Text Processing

  • ✔️ Punctuation Removal: Automatically cleans up subtitle text.
  • ✔️ Smart Lowercasing: Corrects the first letter of each subtitle line.
  • ✔️ Greek Stop Word Handling: Skips common words for efficiency.

• Local AI Integration

  • 🧠 Offline Spell Checking: Integrates seamlessly with a local Ollama server, ensuring your data remains private.
  • 🧠 Context-Aware Corrections: Uses a Greek-language model (ilsp/meltemi-instruct-v1.5) for better corrections.

• Comprehensive SRT Workflow

  • ▶️ Effortless Parsing: Automatically reads and structures .srt files.
  • ✏️ Interactive Editing: Provides an intuitive UI to manually edit corrected text.
  • 💾 Standard Export: Downloads the final, corrected subtitles in valid .srt format.

• User-Friendly Interface

  • ⚙️ Customizable Options: Easily toggle spell checking, punctuation removal, and lowercasing.
  • ⏱️ Real-time Metrics: Tracks processing progress and estimates time remaining.

---

⚙️ Installation

1️⃣ Clone the repository

git clone https://github.com/helenmand/greek-srt-corrector.git
cd greek-srt-corrector

2️⃣ Install dependencies

pip install -r requirements.txt

3️⃣ Install & Run Ollama

Follow the official Ollama installation guide. After installation, pull the required Greek model:

ollama pull ilsp/meltemi-instruct-v1.5

4️⃣ Run the app

streamlit run srt_corrector.py

---

⚙️ Why Meltemi?

The application suggests using ilsp/meltemi-instruct-v1.5 model because it offers a balance of performance and accuracy for Greek-specific tasks.

While larger, more general-purpose models like gemma3:12b may have a higher overall quality rating, the Meltemi model is:

• Lightweight: At 7 billion parameters, it requires significantly less VRAM, making the app more accessible.
• Highly Effective: Its training data is tailored for the Greek language, ensuring high-quality, task-specific corrections. Additionally, the application's use of a user-provided context significantly improves the model's performance by tailoring its corrections to the specific subject matter of the text.

You are free to change the model from the app's sidebar to explore other options that might better suit your needs.

---

📌 Example

The app can automatically correct common spelling mistakes while applying your chosen formatting options.

Input:

1
00:00:02,000 --> 00:00:04,000
<b>Γειατί ήρθες εδώ;</b>

Output: (with spell check, punctuation removal and lowercase options enabled)

1
00:00:02,000 --> 00:00:04,000
<b>γιατί ήρθες εδώ</b>

Correction: The word "Γειατί" was changed to "γιατί", the punctuation mark was removed and the first letter became lowercase