This is a demo repository for my dissertation: "A WHISKER OF TRUTH: A DEEP LEARNING PIPELINE FOR AGE PREDICTION FROM VOCALISATIONS OF THE DOMESTIC FELINE"

Basic UI for demonstration purposes. This tool is designed for predicting the age group (kitten, senior) of samples based on pre-extracted embeddings. The model is trained on a full dataset, with a few samples held back for performance demonstration.

In a production environment, support for uploading wav files to extract embeddings in real-time will be added.

A demo video is included to showcase the functionality of the tool. The actual class column is included for performance demonstration. In a production environment, this column would not be present. Best viewed with sound on for commentary.

The model build (cat_age_model.keras) is packaged and included in this repo. The StandardScaler scaled on the training data, is included as well (scaler_full.pkl). Embeddings will be scaled to expected format after which predictions are made.

To run the demo:

1. Start the application by running binary-demo-app.py
2. Navigate to the port http://127.0.0.1:5001
3. Select "Choose File" to upload a CSV file with embeddings; samples are included in this repo
4. Click on "Predict" to view the predictions and probabilities for each sample