AutoTranscript GUI 🎙️

AutoTranscript is a powerful, GPU-accelerated subtitle generator built on top of OpenAI's Whisper model. It features both a command-line interface (CLI) and a beautiful CustomTkinter-based GUI for users who prefer a graphical workflow.

Supports:

• Local audio/video files
• Subtitle translation to English
• OpenAI API (for higher quality translations)

---

✨ Features

• 🖥️ Full-featured GUI with progress tracking, real-time logs, and OpenAI config
• 📜 Generate .srt subtitle files from media files
• 🌍 Supports multilingual transcription and optional translation to English
• 🧠 Uses Faster-Whisper for fast GPU-accelerated transcription
• 🔁 Automatic model selection based on VRAM (e.g. large-v3, medium, etc.)
• 🔐 API key manager for OpenAI GPT models

---

📸 GUI Preview

---

🧩 Requirements

• Python 3.8+
• ffmpeg (must be installed)
• NVIDIA GPU with CUDA (recommended)
• Whisper models (via Faster-Whisper)
• PyTorch with CUDA
• .env file for OpenAI (optional)

---

📦 Installation

git clone https://github.com/jjaruna/autoTranscriptGUI.git
cd autoTranscriptGUI
pip install -r requirements.txt

---

🚀 Launch the GUI

python AutoTranscriptGUI.py

Whisper Model Comparison

Model	Recommended VRAM	Performance	Use Case
tiny	≥ 1 GB	Very fast, low accuracy	Quick tests, very low-resource machines
base	≥ 2 GB	Fast, low-medium accuracy	Basic transcriptions, short files
small	≥ 4 GB	Balanced speed/accuracy	Good for medium-length files, better accuracy
medium	≥ 8 GB	Slower, higher accuracy	Longer files, good balance of quality and performance
large-v1	≥ 10 GB	High accuracy	Older large model, still very capable
large-v2	≥ 10 GB	Improved accuracy	More robust than v1, slower on limited VRAM
large-v3	≥ 12 GB	Latest model, high accuracy	Best offline model for quality transcription
large-v3-turbo	≥ 12 GB	Fastest large model	High speed with high accuracy, better multi-language support

🧠 Recommendation

After testing the large-v3-turbo model more than 10 times, I can confidently say it is the fastest and most accurate among all Whisper models included in this app.

🖥️ My system has 4GB of VRAM, and despite being under the recommended VRAM for large models, large-v3-turbo still performed exceptionally well.

⚠️ Note: Your experience may vary depending on your GPU and available VRAM. Use this recommendation as a reference, not a guarantee. If you encounter performance issues, try smaller models like medium or small.

---

⚙️ OpenAI API Setup (Optional)

To enable OpenAI-powered translation:

1. Click "Add API Key" in the GUI
2. Enter your OpenAI key and model (gpt-4, gpt-3.5-turbo, etc.)
3. It will be saved to .env file automatically

---

🖥️ CLI Mode (Optional)

You can still use the command-line version via autosub.py:

python autosub.py myvideo.mp4 -l ja --translate --model base

CLI Options

Option	Description
filename	File path
-l, --language	Force language (e.g. en, es, zh)
-t, --translate	Translate to English
-o, --openai	Use OpenAI API
--model	Whisper model to use
--debug	Enable debug mode
--keep	Keep intermediate WAV file

---

📝 Output

• Subtitles are saved as .srt files in the same folder as your media.
• If translated, original and translated text will be preserved.

---

🧪 Example GUI Workflow

1. Open GUI
2. Select video/audio file
3. Choose language and Whisper model
4. (Optional) Enable "Translate to English"
5. (Optional) Enable "Use OpenAI"
6. Click Start Transcription
7. Wait for progress bar and logs to finish

---

🙏 Credits

• Built with OpenAI Whisper
• Powered by Faster-Whisper
• GUI built with CustomTkinter
• Thank you General Koi, for the great help in testing and reviewing the Japanese transcripts.

---

📄 License

MIT License — free for personal and commercial use.