Deep Learning Based Automated Lip Reading for Deaf

Speech recognition systems play an integral role in numerous applications, from virtual assistants to accessibility tools. This paper offers a new viewpoint to speech recognition utilizing computer vision and deep learning techniques. The proposed system is trained on a sizable dataset comprising 700 video clips of individuals uttering predefined words, add up to approximately 3 GB of data. Leveraging TensorFlow and Keras, model architecture is designed incorporating convolutional and dense layers. The training process yielded promising results, with a training accuracy of 95.7% and a validation accuracy of 98.5%, indicative of robust classification performance. The integration of computer vision enriches the system's ability to extract meaningful features from audio-visual inputs, enhancing its overall recognition accuracy. Proposed method demonstrates significant potential for real-time speech recognition applications in different areas, like human-computer interaction and assistive technologies.