🎵 Music Genre Classification 🎶

A deep learning-based system to classify music genres using SVM, CNN, LSTM, and Transformer models, with YAMNet-based feature extraction.

🚀 Features

✅ Extracts audio features using YAMNet embeddings.
✅ Trains multiple models (SVM, CNN, LSTM, Transformer) for comparative analysis.
✅ Evaluates model performance using accuracy, confusion matrices, and classification reports.
✅ Allows users to upload an audio file for real-time genre classification.

---

🎵 What is YAMNet?

YAMNet is a pre-trained deep learning model developed by Google for audio classification. It is based on MobileNet architecture and is trained on AudioSet, a large-scale dataset of sounds. YAMNet extracts high-level embeddings from audio, which can be used for various machine learning tasks, including music genre classification.

🎯 Why YAMNet?

• 🟢 Pre-trained on large-scale data (AudioSet)
• 🟢 Extracts meaningful sound features
• 🟢 Lightweight and efficient for real-time applications

📌 Using YAMNet for Feature Extraction

To extract features from an audio file using YAMNet in Python:

import tensorflow_hub as hub
import numpy as np
import librosa

# Load the YAMNet model
yamnet_model = hub.load('https://tfhub.dev/google/yamnet/1')

# Load and preprocess the audio file
def extract_yamnet_features(audio_path):
    waveform, sr = librosa.load(audio_path, sr=16000)  # Ensure 16kHz sample rate
    waveform = np.reshape(waveform, (1, -1))  # Reshape to match model input
    
    # Run YAMNet model
    embeddings, scores, spectrogram = yamnet_model(waveform)
    return embeddings.numpy()

# Example usage
features = extract_yamnet_features('test_music(hiphop).wav')
print(features.shape)  # Output shape: (number of frames, 1024)

🔗 Reference: TensorFlow YAMNet Tutorial

---

📂 Project Structure

🌆 Music Genre Classification
 ├── 📂 Data
 │   ├── 📂 genres_original        # Raw music dataset organized by genre
 │   ├── 📂 images_original        # Spectrogram images (if applicable)
 │   ├── 📝 music_features.csv     # Extracted features dataset
 │   ├── 📝 test_music(hiphop).wav # Sample test audio file
 │   ├── 📝 yamnet_features.csv    # YAMNet feature embeddings
 │
 ├── 📂 Models
 │   ├── 📝 cnn_model.h5           # Trained CNN model
 │   ├── 📝 lstm_model.h5          # Trained LSTM model
 │   ├── 📝 svm_model.pkl          # Trained SVM model
 │   ├── 📝 transformer_model.h5   # Trained Transformer model
 │
 ├── 📝 .gitattributes             # Git LFS tracking for large files
 ├── 📝 requirements.txt           # Required dependencies
 ├── 📝 FeatureExtraction.ipynb    # Extracts YAMNet features from audio
 ├── 📝 MusicGenreModels.ipynb     # Trains all ML models (SVM, CNN, LSTM, Transformer)
 ├── 📝 Result_Analysis.ipynb      # Evaluates model performance & visualization

---

🛠️ Setup & Installation

1⃣ Clone the Repository

git clone https://github.com/JaspreetSingh-exe/Music-Genre-Classification.git
cd Music-Genre-Classification

2⃣ Install Dependencies

pip install -r requirements.txt

3⃣ Download & Install Git LFS (For Large Model Files)

git lfs install
git lfs pull

---

🎵 How to Run the Project

1⃣ Extract Features from Audio

Run:

jupyter notebook FeatureExtraction.ipynb

This extracts YAMNet embeddings from the dataset.

2⃣ Train Models

Run:

jupyter notebook MusicGenreModels.ipynb

This trains SVM, CNN, LSTM, Transformer models and saves them in the Models/ directory.

3⃣ Evaluate Model Performance

Run:

jupyter notebook Result_Analysis.ipynb

This generates accuracy scores, confusion matrices, and classification reports.

4⃣ Predict Genre for a New Audio File

Run in Python:

from MusicGenreModels import predict_audio_file
predict_audio_file("path/to/audio.wav", models, scaler, label_encoder)

---

📊 Model Comparison

Model	Accuracy
SVM	86.11% (best)
CNN	80.56% (overfitting)
LSTM	71.67%
Transformer	33.33%

---

📌 Future Improvements

• ✅ Deploy as a Web App where users can upload songs & get genre predictions.
• ✅ Improve accuracy using larger datasets & better preprocessing.
• ✅ Experiment with other pre-trained models for better feature extraction.

---

🐜 License

This project is licensed under the MIT License.
You are free to use, modify, and distribute this software as long as the original license is included.

📄 See the LICENSE file for more details.

---

📞 Contact

💡 Author: Jaspreet Singh
📞 Email: [email protected]
🔗 GitHub: JaspreetSingh-exe

---