Sign Language Translator: A Comparative Analysis of Deep Learning and Machine Learning Models

Worldwide, deaf mute people interaction with people who do not know the sign language is quite hard. Even though gestures constitute the main source of communication of the deaf and the mute, many hearing people do not have the ability to read normal signs. To minimize this shortcoming, the Sign Language Translator intends to utilize advanced technologies including deep learning, computer vision as well as machine learning to convert sign language to either oral speech or written text. This system performs the instantaneous conversion of sign language into text thereby promoting contacts and the free interaction of disabled and able people. Our approach is not only arrived at from implanting existing methods on a custom dataset but also uses a larger and richer dataset, WLASL (Word-Level American Sign Language) thus enhancing model performance. Four machine learning models are borrowed in this project: Long Short-Term Memory (LSTM) captures and recalls information over longer periods, Convolution Neural Networks (CNN)samples and analysis parts of images to retrieve spatial information, Random Forest Classifier implements learning using models based on a decision tree, and Support Vector Machine (SVM) is a supervised learning technique used to classify data of high dimensionality. With these models in place, we seek to improve the precision and utility of sign language interpretation.