---
license: mit
---
# Create dataset from video 

This project contains a script based on an OpenAI model that will help you automatically create your own text dataset from video files or podcasts.

Under the hood, it uses the powerful **Whisper-Large-V3** model, which can accept the following audio and video formats: .mp3, .m4a, .wav, .mp4, .mkv, and .webm.

*(We recommend using the .mp3 and .mp4 formats for best compatibility and performance.)*

### Installation and Run

1. Make sure you have Python 3.9 or later installed.
2. Install FFmpeg (required, as extracting audio from video won't work without it).
* On Windows, open a command prompt and enter:
`winget install ffmpeg`
3. Install PyTorch with support for your graphics card (CUDA) for faster recognition:
```bash
pip install torch torchvision torchaudio --index-url [https://download.pytorch.org/whl/cu121](https://download.pytorch.org/whl/cu121)

4. Install the necessary libraries for working with the model:
pip install transformers accelerate

5. Create two empty folders in the script folder:
* `Videos` (place your videos here)
* `model` (for local storage of Whisper weights).

6. Run the script. All processed dialogs will be automatically saved to a clean .txt file.

**⚠️ Licensing Information:**
* The code portion of this project (automation scripts) is distributed under the MIT license.
* The Whisper neural network model itself, developed by OpenAI, is distributed under the Apache 2.0 license.