Deepfake Audio Detection via MFCC Features and VGG16 + LSTM

This project implements the research paper:
"Deepfake Audio Detection via MFCC Features Using Machine Learning"

---

📌 Summary

We detect audio deepfakes using classical ML models and a hybrid deep learning model. The project covers:

• 🧹 Data Cleaning & Preprocessing
• 🎵 MFCC and Spectral Feature Extraction
• ⚙️ Dimensionality Reduction via PCA
• 🧠 ML Models: SVM, RF, MLP, Gradient Boosting
• 🤖 Deep Learning: VGG16 + LSTM Fusion
• 🔍 Hyperparameter tuning with RandomizedSearchCV

---

📂 Dataset Used

Fake-or-Real (FoR):
Includes 4 subsets:

• for-original
• for-2sec
• for-norm
• for-rerec

---

📈 Results & Observations

• PCA helped reduce training time without losing much accuracy.
• VGG16+LSTM gave best performance for robust detection.
• Evaluation used accuracy, confusion matrix, ROC-AUC.

---

🔗 Notebook

Full code in this notebook: deepfake_audio_detection.ipynb

Also on Kaggle

---

🚀 Tech Stack

• Python, NumPy, Pandas
• Librosa, Matplotlib
• Scikit-learn
• TensorFlow / Keras

---

📦 How to Run

git clone https://github.com/<your-username>/deepfake-audio-detection.git
cd deepfake-audio-detection
jupyter notebook deepfake_audio_detection.ipynb

📊 Results (MLP Classifier on FoR-Rerecorded Subset)

Confusion Matrix

ROC Curve