The need for a sign language to communicate on a daily basis is crucial for people with hearing and speech disabilities to communicate with the rest of the society. This paper presents a robust system for sign language recognition utilizing deep learning techniques, with a focus on CNN and transfer learning models, including VGG19, GoogLeNet, and ResNet. The dataset, collected from Kaggle consisting of images representing hand gestures for alphabets (A–Z) and numbers (1–9), was preprocessed and augmented to optimize model performance. A baseline CNN model was developed, while transfer learning approaches leveraged pre-trained architectures to enhance accuracy and reduce training time. Additionally, a hybrid CNN + LSTM model was implemented to recognize dynamic gestures, capturing both spatial and temporal features. Experimental results indicate high accuracy across all models, with ResNet and VGG19 outperforming the baseline CNN. These findings highlight the potential of deep learning in building efficient, scalable sign language recognition systems, contributing to more inclusive communication technologies.