SONA AI is a machine learning project for detecting emotions from voice recordings. It uses audio features extracted with Librosa and trains a custom neural network to classify emotions such as euphoric, joyfully, sad, and surprised.

This project is built to explore real-time audio classification using Python and deep learning. It includes raw audio data, preprocessing scripts, and trained model files. The goal is to build a system that can identify the emotional tone of a voice input.

• Processes raw audio files into features using Librosa
• Splits data into train, validation, and test sets
• Trains models including logistic regression and random forest
• Saves predictions and evaluation metrics
• Structure ready for real-time or batch emotion classification

SONA AI/ ├── data/ │ ├── raw/ # Original audio and metadata │ ├── processed/ # CSVs with processed features │ ├── features/ # Numpy arrays for each emotion │ └── models/ # Trained model files (.pkl) ├── src/ │ ├── config.py # Project config and paths │ ├── dataset.py # Data loading and splits │ ├── preprocess.py # Audio cleaning and feature extraction │ ├── features.py # Feature engineering │ ├── train.py # Model training script │ └── evaluate.py # Model evaluation script ├── requirements.txt └── README.md

1. Install dependencies

pip install -r requirements.txt

2. Make sure your audio files are in data/raw/audio/

3. Run preprocessing and feature extraction
   python src/preprocess.py

4. Train the model python src/train.py

5. Evaluate results python src/evaluate.py

Notes

Some model files and audio clips are already included. You can replace or expand the dataset by adding new .wav files in the same format.

Future Plans • Add real-time microphone input for live emotion prediction • Try other classifiers like CNN or RNN • Add support for more emotion labels