Embedded Deep Learning to Support Hearing Loss Mobility: In-House Speaking Assistant

Abstract

Hearing-impaired people encounter significant mobility issues in their everyday life as the desire to communicate what they think and need in an increasingly crowded urban society grows, all for social, physiological, safety, and self-esteem reasons. Because of the rapid advancement of technology in the twenty-first century, there are several tools available to aid these individuals with their difficulties and to enable them to clarify their thoughts through a number of devices. All of these technologies, however, are prohibitively expensive and difficult to obtain for these individuals. Many of these assistants require an Internet connection to function properly since they involve extensive data processing on certain Cloud services, which can be a limiting factor because Internet connections are not available everywhere, and not everyone has easy access to them. In this study, we offer a methodology for providing a hearing-impaired person with a speaking assistant that is entirely integrated within the device and does not require an Internet connection, making it a highly economical and portable solution for anyone. Our solution embeds one intelligent Deep Learning model, a text-to-speech model, on a single smart mobile device. This model generates an audio file from the text entered by the user and plays it back to the device’s available output speaker. Our work brings us one step closer to fully self-contained embedded intelligent models that use cutting-edge AI to assist hearing-impaired people with communication.