Machine Learning Approach For Indian Sign Language Recognition Utilizing Bert And Lstm Models

Sign language is a visual form of communication that conveys meaning through body language,facial expressions and hand gestures. Language barriers prevent people who don’t sign from interacting with those who do. This is the root of the issue. To improve communication this can be fixed by developing real-time sign language recognition systems using cutting-edge methods. This work presents a hybrid BERT + LSTM model machine learning approach for word-level recognition in Indian Sign Language (ISL). In order to overcome the difficulties in capturing both temporal and spatial features in ISL gestures this model combines the strength of BERTs bidirectional encoder representations with the adaptability of LSTM to handle sequential dependencies in the integration way like proposed BERT+LSTM. To ensure robustness the ISL-Express dataset is made up of a variety of hand gesture images labeled with corresponding ISL words that were recorded under a range of conditions. Regarding recall accuracy precision and real-time processing metrics the results show that the suggested BERT + LSTM model outperforms these alternatives. It specifically achieves a maximum accuracy of 95 % with lower latency and higher frame rates. When contrasted with conventional methods real-time ISL recognition applications can greatly benefit from the models sophisticated performance features. Ultimately this suggested BERT+LSTM model which had been enhanced with data augmentation and regularization techniques was compared to several alternative machine learning algorithms including CNN + LSTM RNN + GRU Transformers + GRU and BERT + GRU.