🎧 Predictify - Urban Sound Classification with Machine Learning

Predictify is a full-stack AI-powered web application that identifies types of urban sounds (like sirens, dog barks, or drilling) from audio files using a trained machine learning model. Users can upload an audio clip through a modern UI, and the app will analyze and predict the sound type using Python and a Random Forest model.

---

🌟 Key Features

• 🎙 Upload and analyze audio files (MP3, WAV, etc.)
• 🤖 Real-time machine learning prediction using Python
• 🧠 Trained with UrbanSound8K dataset
• 💡 Dual feature extraction: MFCC + Mel Spectrogram
• 🕹 Interactive and animated frontend with React + Lottie
• 🌙 Light/Dark mode toggle
• 📁 Audio upload history and playback

---

🛠 Tech Stack

Layer	Technology
Frontend	React, Framer Motion, Lottie
Backend	Node.js, Express, Multer
ML Model	Python, Librosa, Scikit-learn
Dataset	UrbanSound8K

---

🚀 Quick Start

1. Clone the repository

git clone https://github.com/mezone1994/predictify.git
cd predictify

2. Setup Backend

cd Backend
npm install
node server.js

3. Setup Frontend

cd ../client
npm install
npm start

4. Train ML Model

cd ../
python preprocessing/preprocessing.py

---

📁 Project Structure

predictify/
│
├── Backend/                # Node.js server files
│   ├── controllers/        # Controller logic
│   ├── models/             # Python execution logic
│   ├── routes/             # Express routes
│   └── server.js           # Main server file
│
├── client/                 # React frontend
│   ├── App.js              # Main React app
│   └── TeamModal.js        # Team member info modal
│
├── preprocessing/          # Feature extraction and model training
│   └── preprocessing.py    # Full training script
│
├── uploads/                # Uploaded audio files
├── deploy.py               # Predicts class of uploaded file
├── voice_processing_model.pkl  # Saved ML model
├── *.csv                   # Processed training/validation/test data
├── UrbanSound8K/           # Dataset and metadata
└── README.md

---

🧪 How Prediction Works

1. Audio file is uploaded via React and sent to Express.
2. File is saved in /uploads using Multer.
3. deploy.py is executed from Node.js using child_process.spawn().
4. Python:
   • Loads the .pkl model
   • Extracts MFCC features
   • Predicts class using Random Forest
   • Returns class ID, label, and description
5. Node parses result and sends it to the React frontend.

---

🧠 Model Training Overview

Step-by-step (in preprocessing.py):

• Extract MFCC or Mel Spectrogram from UrbanSound8K.
• Save datasets as:
  • training_file_method1.csv
  • validation_file_method1.csv
  • testing_file_method1.csv
• Train two Random Forest models (one for each method).
• Evaluate and compare them.
• Save the best performing model as voice_processing_model.pkl.

---

🔉 Example Sound Classes

ID	Label	Description
0	air_conditioner	Humming of an AC unit
1	car_horn	Beep sound of a car horn
2	children_playing	Laughter and playful noises from kids
3	dog_bark	Woof or yap from a dog
4	drilling	Mechanical sound of a drill
5	engine_idling	Car engine running while parked
6	gun_shot	Loud sharp sound of a firearm
7	jackhammer	Percussive drilling noise
8	siren	Emergency siren wail
9	street_music	Outdoor music performance

---

📷 UI Preview

---

📚 References

• UrbanSound8K Dataset
• Librosa
• Scikit-learn

---

👨‍💻 Developed By

*Group 13 - Mcmaster Students

---

📄 License

MIT License © 2025 Moayed Mahmoud
Free to use and modify with credit.