Voice-Based Stress Detection

Overview

This project detects stress in speech audio using MFCC and time-series acoustic features.

Dataset used:
RAVDESS (Ryerson Audio-Visual Database of Emotional Speech and Song)

We convert emotion labels into:

• Normal → neutral (01), calm (02)
• Stressed → angry (05), fearful (06)

Project Structure

VoiceStressProject/
│
├── dataset/
│ ├── normal/
│ ├── stressed/
│
├── stress_pipeline.py
├── requirements.txt
└── README.md

Features Extracted

• 13 MFCC coefficients (mean + variance)
• Zero Crossing Rate
• Spectral Centroid
• Energy
• Pitch (mean + variance)

Total features per sample: 31

Model

RandomForest Classifier

Evaluation Metrics:

• Accuracy
• Precision
• Recall
• Confusion Matrix

How to Run

1. Install dependencies:

   pip install -r requirements.txt

2. Run the pipeline:

   python stress_pipeline.py

Output

• Feature matrix shape
• Accuracy score
• Classification report
• Confusion matrix