Video Transcriber and Translator

A Python script that automates video subtitle creation & translation, supporting both local files and online video URLs.

🚀 Features

Download videos from various platforms easily by specifying the URL (YouTube, Twitter, Facebook, Instagram, etc.)
Transcribe audio using ChatGPT's Whisper model (default) or local Whisper model
Translate the generated transcriptions to a target language using OpenAI's GPT models
Add translated subtitles to videos
Support for multiple languages and resolutions (including YouTube Shorts)
Support for long videos, these are split into chunks automatically before transcribing and translating

🎬 Demo

Here are some examples of the script in action:

1. English to Korean Translation

2. English to Spanish Translation

3. German to English Translation

4. Spanish to English Translation

Note: Results may vary with accents or background music, especially when using the local model. Video and caption synchronization might be affected when the audio isn't clear. The above clips were processed using OpenAI for both transcription & translation (default script behavior).

🖥️ Usage

Run the script using the following command:

python translate.py video_input target_language [options]

Arguments:

video_input: URL or path to the input video file
target_language: Target language for translation (e.g., Spanish, English, French)

Options:

--output_dir: Directory to save output files (default: "output")
--models_path: Path to store Whisper models (default: "Models")
--openai_api_key: OpenAI API key (if not set as an environment variable)
--font: Font to use for subtitles (default: "NanumGothic")
--use_local_whisper: Use local Whisper model for transcription instead of ChatGPT's Whisper

📋 Examples

Translate YouTube video subtitles to Spanish (using default ChatGPT Whisper):
```
python translate.py https://www.youtube.com/watch?v=VIDEO_ID Spanish
```
Translate local video file subtitles to French (using default ChatGPT Whisper):
```
python translate.py /path/to/your/video.mp4 French
```

Use a specific output directory and font:

python translate.py input_video.mp4 German --output_dir my_output --font Arial

Use a local model for transcription:

python translate.py input_video.mp4 Korean --use_local_whisper

🛠️ Installation

1. FFmpeg Installation

Windows:

Download FFmpeg from ffmpeg.org
Extract the ZIP file
Add the bin folder path to system PATH
Verify installation: ffmpeg -version

macOS:

brew install ffmpeg

Linux:

sudo apt update
sudo apt install ffmpeg

2. Python Dependencies

Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate  # Linux/macOS
venv\Scripts\activate     # Windows

Install required packages:

pip install openai==1.12.0
pip install faster-whisper==0.10.0
pip install yt-dlp==2024.3.10
pip install ffmpeg-python==0.2.0
pip install pydub==0.25.1

3. OpenAI API Setup

Create account at platform.openai.com
Generate API key in account settings

Set environment variable:

# Linux/macOS
export OPENAI_API_KEY='your-key-here'

# Windows (PowerShell)
$env:OPENAI_API_KEY='your-key-here'

4. Download the repository

Clone repository:

git clone https://github.com/tikene/video-caption-and-translate.git
cd video-caption-and-translate

Verify installation:
```
python translate.py --help
```

⏳ Common Issues

FFmpeg Not Found

Ensure FFmpeg is in system PATH
Restart terminal/IDE after PATH changes
Check with ffmpeg -version

OpenAI API Errors

Verify API key is set correctly
Check account has sufficient credits
Ensure stable internet connection

📂 Output

The script generates the following files in the output directory:

Downloaded video (if URL was provided)
Translated SRT subtitle file
Video with embedded translated subtitles

⚠️ Important Notes

The script uses the GPT-4 model for translation by default, which costs around $0.1 cents for a two-minute video. You may reduce token costs by switching to 'gpt-4o-mini' which is ~80% cheaper at the expense of translation quality
Longer video → Higher costs (duh)

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.