Daily Life Conversation in Bangla Sign Language Using Machine Learning Technique

Sign language is a way of communicating and interacting for access to information. Currently, deaf and mute people rely heavily on sign language to communicate. It enables the challenged person to interact with a hearing-enabled person to express themselves and participate in all facets of life. In these circumstances, research is needed in the area of sign language recognition as limited work has been conducted. Also, as it varies from region to region and in the context of BangIa sign language, related research is more mandated. The dataset in this study consisted of 9,000 video data. Alongside this, Artificial Neural Networks (ANN) and Long Short-Term Memory (LSTM) were used to establish a two-way communication system. LSTM is used for long-term dependencies and sequence predictions. To gather video data and extract characteristics from the video frames, MediaPipe and OpenCV were employed. Fine-tuning techniques, such as the number of hidden layers, the density of layers, the number of neurons within each layer, the learning rate, the optimizer, and other parameters were used to determine the optimal accuracy. The results showed that 96% of the categories were accurately represented by scikit-Iearn. This result offers a reference for employing deep learning structures for the detection of BangIa sign language and the development of dependable assistive technology tailored to the necessity of the hearing-impaired community.