🎵 Music Genre Classification System

This project is an end-to-end Music Genre Classification system that uses machine learning / deep learning techniques to automatically classify audio tracks into their respective music genres based on extracted audio features.

The goal of the project is to demonstrate how signal processing + machine learning can be combined to solve a real-world audio classification problem.

---

🚀 Project Motivation

With the massive growth of digital music platforms, organizing and recommending music manually has become impractical.
Music genre classification helps in:

• 🎧 Music recommendation systems
• 📂 Automatic music library organization
• 🎼 Music information retrieval (MIR)
• 🤖 Intelligent audio-based applications

This project focuses on building a clean, interpretable, and effective genre classification pipeline rather than using black-box APIs.

---

🧠 How the System Works

1️⃣ Audio Processing

• Raw audio files are loaded from the dataset
• Each audio file is split into manageable segments (if required)
• Audio signals are converted into numerical representations using feature extraction

---

2️⃣ Feature Extraction

The following audio features are extracted from each track:

• MFCCs (Mel-Frequency Cepstral Coefficients)
• Chroma Features
• Spectral Centroid
• Spectral Bandwidth
• Zero Crossing Rate
• RMS Energy

These features capture:

• Timbre
• Pitch distribution
• Frequency characteristics
• Rhythm and energy

📌 Feature extraction converts raw sound waves into a format suitable for ML models.

---

3️⃣ Model Training

• Extracted features are used as input to the classification model
• The model learns patterns that differentiate music genres
• The trained model predicts the genre of unseen audio samples

---

4️⃣ Genre Prediction

• For a new audio file:
  • Features are extracted
  • Passed through the trained model
  • The most probable genre is returned as output

---

🏗️ Model Architecture

Depending on implementation, the project uses:

🔹 Machine Learning Approach

• Classifier: Random Forest / SVM / Logistic Regression
• Well-suited for structured feature vectors
• Easy to interpret and fast to train

🔹 Deep Learning Approach (if used)

• Neural Network (Dense / CNN)
• Learns complex feature interactions
• Better performance on large datasets

📌 Why Feature-Based Models?

• Computationally efficient
• Works well with limited datasets
• Easier to debug and explain than raw-audio deep models

---

🧪 Dataset

• Dataset Used: GTZAN Music Genre Dataset
• Number of Genres: 10
  • Blues
  • Classical
  • Country
  • Disco
  • Hip-Hop
  • Jazz
  • Metal
  • Pop
  • Reggae
  • Rock

Preprocessing Steps:

• Audio normalization
• Fixed-duration sampling
• Feature scaling (Standardization)

---

🖥️ Tech Stack

Core Technologies

• Python
• NumPy
• Pandas
• Librosa (Audio Processing)
• Scikit-learn
• TensorFlow / Keras (if deep learning used)
• Matplotlib / Seaborn (Visualization)

---

📊 Evaluation Metrics

The model performance is evaluated using:

• Accuracy
• Precision
• Recall
• F1-score
• Confusion Matrix

These metrics help analyze:

• Genre-wise performance
• Misclassification patterns
• Model reliability

---

📁 Project Structure

music-genre-classification/
│
├── data/
│   └── genres_original/
│
├── notebooks/
│   └── music_genre_classification.ipynb
│
├── models/
│   └── trained_model.pkl
│
├── utils/
│   └── feature_extraction.py
│
├── requirements.txt
└── README.md

⚙️ Setup Instructions

1️⃣ Clone the Repository

git clone [https://github.com/your-username/music-genre-classification.git](https://github.com/your-username/music-genre-classification.git)
cd music-genre-classification

2️⃣ Create Virtual Environment (Recommended)

python -m venv venv

Activate Environment

Linux / macOS:

source venv/bin/activate

Windows:

venv\Scripts\activate

3️⃣ Install Dependencies

pip install -r requirements.txt

4️⃣ Run the Notebook / Script

jupyter notebook

Open the notebook and run cells sequentially to: -> Extract features -> Train the model -> Evaluate results

🧩 Future Improvements

🎙️ Real-time audio genre prediction 🧠 CNNs on Mel-Spectrograms 📱 Web or mobile-based interface 🎶 Multi-label genre classification 🔊 Noise-robust training

📌 Key Learnings

Audio data requires signal processing knowledge Feature quality directly impacts model performance Classical ML models can perform surprisingly well on audio tasks Proper evaluation is critical in multi-class classification problems

🤝 Acknowledgements

GTZAN Dataset Librosa documentation Scikit-learn community

⭐ If you found this project useful, consider starring the repository!