Using GRB voice assessment scale, multimodal deep learning models and multiple outputs for automatic pathological voice quality score 使用GRB嗓音狀況評估量表、多重深度學習模型和多重輸出進行自動病理嗓音品質評分

Mandarin version: xxxx

This database is from Far Eastern Memorial Hospital(FEMH), which is not public for others

The project is in EEA683 course from Yuan Ze University, Taiwan

Content

• Introduction
• Environment
• Motivation
• Required-question
• Feature-extration
• Example
• Reference

Introduction

In this project, it has two types of data. They are "/a voice" and "continuous voice". It has a correspondding GRB voice assessment scale. Here, we use several accoustic features as well as using deep learning method to classify GRB score.

Environment

• Python: 3.6.5
• Keras: 2.2.5
• Tensorflow: 1.14.0

Motivation

In this modern society, a great number of people are suffering voice disorder since it influences their daily life negatively. The dramatic changes in voice may be due to lesions of the vocal cords, such as atrophy of the vocal cords, polyps and tumors, etc. When doctors or professional physicians diagnose patient's vocal folds, patients might suffer from vocal cord examinations, which is time-consuming and may have misjudged situations. Moreover, owing to the fact that the location of the disease is deep in the throat, it needs to be accurately checked through professional instruments such as endoscopes.

In this final project, we will obtain the accousitc features by capturing voice feature signals, through machine learning and deep learning algorithms. The goal is to establish a high-accuracy classification system, and then combine it with the medical field to improve the effectiveness of physicians and solve difficulties in medical diagnosis.

Required-question

Feature-extration

Reference