Because they are unable to convey their thoughts to others who can hear, deaf persons may have a difficult time controlling their emotions and may find that their routines are often interrupted. This is the primary impetus behind the progression of things. These individuals would be helped by this system, and it would make it possible for them to convey their ideas in a manner that is on par with that of persons who do not have any kind of physical impairment. Recent advances in artificial intelligence have made it possible to design a system that is capable of resolving this problem, which was previously thought to be impossible. The purpose of this research is to develop, for the benefit of the deaf community, a system that can translate speech into text. Because it’s possible that some individuals won’t comprehend only the written word, the speech will also be translated into international sign language (ISL). In this method, persons who are deaf or hard of hearing will have their sign language translated into spoken language. In order to develop the most accurate model possible, we will integrate NLP with a variety of ML and AI techniques. For the purpose of prediction, convolutional neural networks (CNN) will be used since lip movements are continuous and rapid, making them challenging to capture. CNN and attention-based long and short-term memory (LSTM) approaches will function since they are both effective at predicting visual input. In order to get more desirable results, we will be using several techniques for augmenting the data. TensorFlow and Keras are going to be the Python libraries that are employed in order to do the voice-to-text translation. At the moment, there is an abundance of software accessible; however, all of them require the use of a network. This piece of hardware can function very well without being connected to the internet. Using the suggested technique, we were able to attain a 100% accuracy rate in sign language prediction and a 96% accuracy rate in phrase understanding.
This is a preview of subscription content, log in via an institution to check access.
