In this study, we propose a deep learning model that recognizes American Sign Language (ASL) in real time and offers a translation both in text and speech using computer vision and a Text-to-Speech (TTS) API. Instead of restricting the dataset to the 26 letters of the English alphabet or a few words, the model uses a dataset containing 20 words and their corresponding gestures in American Sign Language (ASL) to expand the scope and use case of the model. The study introduces a novel architecture that uses Recurrent Neural Network with a larger dataset to train the deep learning model. This model to be used is a Long Short-Term Memory (LSTM) network in collaboration with the Media Pipe framework which will be responsible for feature engineering in order to capture and understand the dynamic nature of sign language. According to the World Health Organization, 430 million people have disabling hearing impairments, and the number has been projected to reach over 700 million by 2050. This research provides a means of communication for these individuals which is a big step towards creating a world where everyone can actively and effectively communicate with one another, no matter how limited their abilities are. This study proposed a model trained on a larger and more complex dataset consisting of 20 dynamic signs while retaining and even surpassing the accuracy of existing models. Two key algorithms are introduced to work simultaneously with the model to ensure smooth and accurate translation when tested in real-time.