Published Paper Details:

Visual Speech Recognition or lip reading is a growing area of research in artificial intelligence and computer vision. The main goal of lip-reading systems is to recognize spoken words by looking at lip movements without using sound. These systems are really helpful in places, surveillance systems, generating subtitles for videos and technology for people who are hard of hearing. This paper presents a lip-reading model that uses learning to turn silent videos into text. The model works by finding the mouth in a video then using a special architecture that combines three types of networks Three-Dimensional Convolutional Neural Networks or 3D-CNN, which look at the video and find patterns in space and time. EfficientNetB0, which helps find the important features in the video. Bidirectional Long Short-Term Memory or Bi-LSTM, networks, which look at how lip movements change over time. The model uses something called Connectionist Temporal Classification loss to predict sequences of words without needing to line up each frame of video with a word. Tests show that the model can recognize speech from videos and produce accurate text. The model can be used in areas, such, as Helping people communicate, Security monitoring, Human-computer interaction, Making multimedia more accessible.