Deep Learning-Based Interpreter For Sign Language

Abstract

Organizational effectiveness is the dearth need in this hyper-competitive era. Survival and competitiveness depend upon effective communication. Communication is acquiring the desired feedback through the meaningful and accurate transference of messages. It is subjective to four core areas including reading, writing, speaking, and listening expertise. It becomes more challenging at the time of learning and transference of knowledge among exceptional individuals. They have commonly practiced sources of sign language available in the given case. Thus, the given project is unique in terms of its exclusive employability for an automated AI-based efficient communication tool as an alternative to sign language. We designed a sign language interpreter that recognizes static signs in American Sign Language (ASL). We made our own dataset to get better results. We used 24 classes in the dataset thus resulting in a total of 24000 images. Then, we trained models with an 80- 20 split, including ResNet50 and InceptionV3 using our dataset. We compared their accuracies and discovered that InceptionV3 produced superior outcomes. We deployed our model on a web page we designed. The program's goal is to close the communication gap between the hearing- and speech-impaired and the rest of society. The web page gives users a quick and effective means to communicate with one another without the use of an interpreter. This can be especially helpful when prompt communication is necessary, like in emergency situations or in the medical field. The webpage is simple to use and intuitive, making a wide spectrum of people able to access it.