Although speech is still the most widely used mode of communication, some persons have difficulties with hearing or speaking. For the kinds of limitations, communication poses a serious challenge. Using deep learning techniques can aid in breaking down obstacles to communication. Here deep learning-based approach that can identify words from gestures is proposed. CNN, LSTM, and GRU (feedback-based learning models), three distinct types of deep learning concepts are utilized to identify signs in Indian Sign Language (ISL) separating video frames. With own dataset, IISL2020, we tested four various hybrids of CNN, GRU, and LSTM. With one LSTM layer and then GRU, CNN which stands the suggested approach achieves around 97% accuracy.