DDMD
DDMD: Digital Drug Music Detector
What it does
The DDMD project aims to provide an AI-powered solution for detecting digital drug music. The process begins with the collection of 3,176 non-copyrighted MP3 files, of which 1,676 are classified as digital drug music and 1,500 as not digital drug music.
These files are used to develop a Random Forest-based model that can classify audio files as either digital drug music or not. The model utilizes 34 features extracted from these audio files, covering both frequency and time-domain characteristics, and has achieved very encouraging results.
To make the results accessible and user-friendly, we developed a web application. This application leverages the trained Random Forest model to classify audio files through a simple and intuitive interface. The DDMD web application is built using Flask for the backend and HTML/CSS for the frontend. It utilizes the pre-trained Random Forest model to make predictions, accepting various audio file formats, converting non-MP3 files to MP3, and allowing input via a YouTube URL.
Additionally, to enhance classification accuracy, we are exploring ways to improve the DDMD application by fine-tuning the Gemini-1.5-flash model. We have proposed two approaches: first, we prepared a JSON dataset based on the Random Forest classifier results to fine-tune the Gemini-1.5-flash model. In the second approach, we are using the Gemini API and experimenting with different input formats, including CSV and JSON, to fine-tune this Gemini model.
Built with
- Web/Chrome
- Google Colab
- Google Slides
- Google Sppech-to-Text AI
Team
From
Algeria