This repository contains code for a real-time emotion detection system using audio data. The system is built using machine learning techniques, including feature extraction, data augmentation, and a pre-trained neural network model.
Real-Time Detection: The system can analyze emotions in real-time from audio input. Machine Learning Model: A pre-trained neural network model is used for accurate emotion prediction. Data Augmentation: The training data is augmented to improve model robustness.
Model Loss - 0.464 when trained on 100 epochs
Predicted Emotions: Anger(1), Disgust(2), Fear(3), Happy(4), Neutral(5), and Sad(6)
-
Preparation:
- Install the required dependencies using
pip install -r requirements.txt. - Place your pre-trained model weights (
pretrained_model_weights.h5) in the project directory. - Update the
model_architecture.jsonfile with the architecture of your pre-trained model.
- Install the required dependencies using
-
Run Real-Time Detection:
- Use the
real_time_detectionfunction in the provided Python script to perform real-time emotion detection on audio input.
- Use the
-
Customization:
- If you want to use your own pre-trained model, make sure to update the model architecture and weights accordingly.
- Adjust feature extraction, data augmentation, or model parameters as needed.
This project utilizes libraries such as Librosa, scikit-learn, and TensorFlow. Special thanks to the contributors of these open-source projects.
Feel free to explore, modify, and integrate this code into your projects. If you encounter issues or have suggestions, please open an issue.