Skip to content

Latest commit

 

History

46 Commits

Folders and files

Repository files navigation

Automate-Grading-Test-NART

An automated tool for grading the National Adult Reading Test (NART) using machine learning to assess cognitive ability through word pronunciation classification.

Build Status
License: MIT

Project Overview

This project focuses on the automated grading of premorbid ability tests, specifically reading tasks measured by the National Adult Reading Test (NART).

The goal is to classify the correct pronunciation of words to assess cognitive ability automatically.


Dataset

This project uses a dataset of audio samples where participants are asked to pronounce a list of 5o words by the National Adult Reading Test (NART). Each sample is labeled as either "correct(0)" or "incorrect(1)" based on the pronunciation accuracy.

Note: The dataset is not publicly available. Please ensure you have the appropriate permissions to use the dataset or collect your own data for testing the model.


Model Details

The model is built using Keras with a Convolutional Neural Network (CNN) architecture.

The input to the model is a Mel-spectrogram of the audio, which represents the frequency spectrum over time.


Model Highlights

Preprocessing: Converts audio to Mel-spectrogram features.

Architecture: RCNN and Transformer layers to capture spatial patterns in the spectrogram.

Evaluation: Metrics such as accuracy, precision, and recall are used to evaluate model performance.


requirements.txt

Keras==2.4.3

tensorflow==2.5.0

numpy==1.19.5

pandas==1.2.3

librosa==0.8.0

Releases

Packages

Used by

Contributors

Languages