Significance of Convolutional Neural Network in View of Lip Reading for Speech-Impaired People

One way to understand what another person is saying is by lip reading. This is done by observing the speaker's lips movements and determining the words being spoken. One of the key problems with lip reading methods is visual uncertainty, making it difficult to implement in practice. When homophones have similar or nearly identical lip movement, this occurs at the word level. ([p], [m], [b] are all examples). One of the primary factors behind this effort is the intention to aid those who struggle with speech problems. Many people throughout the globe have difficulties communicating because of speech impediments, but visual speech recognition has the potential to change that. In the past ten years, as deep learning has developed, there has been a steady increase in the requirement for effective speech-to-text systems. Intelligent devices that can read lips are called lip reading systems. In order to convert lip movement into text, it is necessary to identify the face, find the lips, identify the characteristics, and then train the classifier to detect lip movement. This article presents an extensive examination of the latest advancements in deep learning model-based Visual speech recognition (VSR). The emphasis is on addressing data problems, task-specific complexities like homophones and absence of context, and providing existing results. Progress in these areas will accelerate the implementation of the silent speech interface from concept to reality. In addition to that, we also summarised the statistical results of standard datasets of VSR.