Speech recognition in digital videos without audio using convolutional neural networks

Speech recognition with visual context is a technique that uses digital image processing to detect lip movements within the frames of a video to predict the words uttered by a speaker. Although models with excellent results already exist, most of them are focused on very controlled environments with few speaker interactions. In this work, a new implementation of a model based on Convolutional Neural Networks (CNN) is proposed, taking into account image frames and three models of audio usage throughout spectrograms. The results obtained are very encouraging in the field of automatic speech recognition.