Automatic GRBAS Scoring of Pathological Voices using Deep Learning and a Small Set of Labeled Voice Data

• Authors: Shunsuke Hidaka, Yogaku Lee, Moe Nakanishi, Kohei Wakamiya, Takashi Nakagawa, and Tokihiko, Kaburagi
• Journal: Journal of Voice
• Date: 2022
• Link: https://www.sciencedirect.com/science/article/pii/S0892199722003472

About

This repository contains the implementation of "Automatic GRBAS Scoring of Pathological Voices using Deep Learning and a Small Set of Labeled Voice Data".

Note that some scripts in Supplementary_Programs/ are only available in "Supplementary Materials" of our paper.

Contents

• data/: Dummy dataset (all audio data were identical and uttered by the author, Hidaka).
• notebooks/: Examples of time-frequency representations and data augmentation techniques.
• sample_audio/: Sample audio files used in the notebooks.
• shell/: Shell scripts.
• src/: Python scripts.
• tests/: Test scripts.
• Supplementary_Programs/: Programs for acoustic feature calculation.

Dependencies

Install Poetry if it is not already installed.

pip install poetry

You can install the necessary packages with the following command:

poetry install

Usage

First, activate the virtual environment created by Poetry.
Then, you can run an experiment on the dummy dataset with the following code:

./shell/sample.sh

Note that the dataset is dummy, so the learning will not be successful.