🧠 ManoVani – Conversational AI for Depression Detection

ManoVani is a voice-enabled AI system that intelligently detects signs of depression by analyzing both speech and text. It simulates an empathetic dialogue with users, offering support while classifying their mental state into no depression, mild/false depression, or true depression. Built using machine learning and natural language processing, it aims to provide early mental health intervention.

---

🌟 Features

• 🗣️ Voice Assistant (Sakhi) – Natural conversation via voice
• 📊 NBTree Classifier – Hybrid Naive Bayes + Decision Tree model
• 🔍 Feature Engineering – PCA & standard scaling applied on extracted features
• 🧾 Data Source – DAIC-WOZ dataset for depression detection
• 📁 Audio Feedback – Generates AI voice output after analysis
• 🔒 Privacy-first – Local inference, no personal data stored or uploaded

---

🧠 How It Works

mermaid

    A[User Voice Input] --> B[Speech-to-Text + Preprocessing]
    B --> C[Feature Extraction]
    C --> D[Scaling & PCA]
    D --> E[Ensemble ML Model (NBTree)]
    E --> F[Depression Classification]
    F --> G[Voice Response (response.mp3)]

🗂️ Project Structure

ManoVani/
├── README.md                  # Project overview (this file)
├── index.md                   # GitHub Pages version of README
├── _config.yml                # GitHub Pages theme config
├── main.py                    # Runs full system pipeline
├── sakhi.py                   # Handles voice input/output
├── model.py                   # Loads model and transformers
├── best_ensemble_model.pkl    # Trained depression detection model
├── scaler.pkl                 # StandardScaler for normalization
├── pca.pkl                    # PCA dimensionality reduction
├── processed_data.pkl         # Preprocessed dataset features
├── response.mp3               # AI's voice response
├── requirements.txt           # Python dependencies
└── LICENSE                    # MIT License (recommended)

📦 Installation Clone the repository

git clone https://github.com/NandiniJaiswal05/Fake-Depression-Detection-Using-Speech-Analysis.git
cd Fake-Depression-Detection-Using-Speech-Analysis

Install dependencies

pip install -r requirements.txt

▶️ Run the Application

python main.py

Speak when prompted.

The AI will process your voice and provide a depression assessment.

Response is saved and played from response.mp3.

🧪 Dataset Used

• Name: DAIC-WOZ Dataset

• Files Used: covarep.csv, formant.csv, transcript.csv, audio.wav

• Purpose: Detect depression from speech and facial cues during interviews.

##🔍 Model Details

• Algorithm: Ensemblled Learning (RandomForest, XGBoost, Voting Classifier, NBTree)

Preprocessing:

• StandardScaler for normalization

• PCA for reducing noise and complexity

Performance: Optimized on validation subset of DAIC-WOZ

##🛠️ Tools & Technologies

• 🐍 Python 3.11+

• 🔬 scikit-learn

• 🧠 NLTK, NumPy, pandas

• 🧏 SpeechRecognition, PyAudio, gTTS

• 📊 PCA, StandardScaler

• 📁 Git, GitHub Pages

🌐 GitHub Pages Deployment

Project is also accessible via GitHub Pages: 📍 https://NandiniJaiswal05.github.io/ManoVani

💡 Future Enhancements

• ✅ Multilingual support (Hindi, Marathi)

• ✅ Full frontend (Streamlit or React)

• ✅ Emotion timeline and progress tracking

• ✅ Doctor/counselor integration API

• ✅ Docker & Android deployment

📜 License This project is licensed under the MIT License.

⚠️ Disclaimer: This is a research-based assistant and not a substitute for professional mental healthcare.

🙋‍♀️ Author Nandini Jaiswal AI/ML Developer | Mental Health Enthusiast 📍 GHRCE, India 🔗 LinkedIn

🤝 Contributions Pull requests, feature ideas, and bug reports are welcome!

# Fork → Code → Pull Request ✔️

📬 Contact 📧 nandinijaiswal783@gmail.com 📦 GitHub: NandiniJaiswal05