🧠 Speaking Disorder Detection Using Convolutional Neural Networks (CNN)

A deep learning solution to identify speaking disorders, specifically dysarthria, using audio feature extraction and 1D CNNs.

---

📌 Project Overview

This project detects dysarthric speech using Mel-Frequency Cepstral Coefficients (MFCC) and other spectral audio features. The model classifies .wav files as either normal or disordered speech with high accuracy.

---

📂 Folder Structure

├── All Dys Wav/             # 🧑‍⚕️ Dysarthric audio samples (label: 1)
├── All Non Dys Wav/         # 🙂 Normal speech samples (label: 0)
├── Testing Dataset/         # 🧪 Additional unseen test samples
├── images/                  # 🖼️ Contains model diagram and screenshots
│   └── img1.png
├── Dysarthria_Detection.ipynb   # 🧠 Main training and evaluation notebook
├── README.md                # 📘 You are here

---

🔍 Feature Extraction

Speech files are processed using librosa to extract:

• MFCC (Mel-Frequency Cepstral Coefficients)
• Chroma Features
• Spectral Contrast
• Tonnetz
• Zero-Crossing Rate

---

🧠 CNN Model Architecture

Layer Type	Description
Conv1D	Extracts patterns from MFCCs
MaxPooling1D	Downsamples feature maps
Flatten	Flattens output for dense layers
Dense + Dropout	Fully connected layers to prevent overfitting
Softmax	Final classification output layer

---

🧪 Sample Output

Input: dys_001.wav
Predicted Class: 1 (Dysarthric Speech)
Confidence: 93.4%

---

▶️ How to Run

1️⃣ Clone the repository

git clone https://github.com/Saim-Nadeem/Speaking-Disorder-Detection-Using-Convolutional-Neural-Networks-CNN-.git
cd Speaking-Disorder-Detection-Using-Convolutional-Neural-Networks-CNN-

2️⃣ Install dependencies

pip install -r requirements.txt

3️⃣ Launch Jupyter Notebook

jupyter notebook Dysarthria_Detection.ipynb

4️⃣ Run all cells

• Upload your test .wav files in the Testing Dataset/ folder
• Watch the model predict normal or disordered speech

---

🛠 Requirements

• numpy
• pandas
• librosa
• scikit-learn
• tensorflow
• absl-py
• keras

---

🖼 Visual Overview

---

👤 Author

Saim Nadeem
🔗 GitHub: Saim-Nadeem