Voice Restoration Device Using Machine Learning of Acoustic and Visual Output During Electrolarynx Use;Conquering Obstacles to the Integration of Diagnostic AI in Healthcare

Agrawal, Sameer, School of Engineering and Applied Science, University of Virginia
Dong, Haibo, University of Virginia
Ferguson, Sean, EN-Engineering and Society, University of Virginia

Technological unemployment, technology replacing jobs previously held by humans, is an issue that has created fears against the implementation of innovative technologies within various workspaces. Common examples of this theme include the replacement of retail cashiers with self-service checkout kiosks and the displacement of manual labor with robotic equipment in manufacturing. Although automation has historically been limited to low-skill jobs, more recently, this trend has seen an increase in automation within specialty fields such as medicine and healthcare through machine learning and artificial intelligence (AI). As such, the proposed technical and STS research paper focus on how AI will shape medicine in the coming years and the barriers to implementing this technology.
The technical paper focuses on using machine learning and artificial intelligence to create an algorithm that improves speech for electrolarynx users. We hypothesize that a trained artificial neural network (ANN) which consists of a combined Convolutional Neural Network (ConvNet) and Long Short-Term Memory (LSTM) network can interpret the visual and auditory signal of an electrolarynx user, and output a computer generated, intelligible, “normal” voice. The articulatory lip movements of speech and the acoustic output from the electrolarynx will serve as the known input data for the machine learning model. The laryngeal speech will be the known output response for the machine learning model. The articulatory movement and acoustic speech data will be processed by a customized ANN classifier. A unique ANN classifier for each phoneme in the English Language will be created to simulate laryngeal speech.
The STS paper is focused on the barriers to implementing AI to the healthcare system. In order to analyze how the healthcare system is impacted by the incorporation of Artificial Intelligence both on a macro and micro level, The actor network theory (ANT), thus provides a framework that analyzes the interplay between the different actors, be they human or non-human, as is the case with the artifact of machine learning. There have been several concerns raised as to the adoption of AI within medicine, many of them centered around the attitudes of medical care professionals. These include beliefs that the technology could affect professional autonomy while diagnosing or treating patients, that it may be used as a means of control by hospital administrators and that it may interfere in the relationships between medical professionals and patients.

BS (Bachelor of Science)

School of Engineering and Applied Science

Bachelor of Science in Biomedical Engineering

Technical Advisor: Haibo Dong

STS Advisor: Sean Ferguson

Technical Team Members: Surabhi Ghatti, Medhini Rachamallu, Katherine Taylor

All rights reserved (no additional license for public reuse)
Issued Date: