Lip2Voice: a sequence-to-sequence visual speech recognition system for predicting speech from silent video inputs

Humans have always paid high attention to lip movements for visually listening to speech in a noisy environment [1].Furthermore, speech can be interpreted without auditory input.This may be required when audio recognition is complex, in cases of speech recovery during online video sessions or from corrupted videos, in military applications, and closed-circuit television (CCTV) surveillance.Advancements in computer vision have led to research efforts aimed at automating the process of lipreading, known as the visual speech recognition (VSR) system [2].VSR primarily involves the analysis of visual aspects of speech.VSR systems aim to equip machines to comprehend speech even in noisy environments.Speech recognition systems can be modelled using audio-visual data, audio samples, and silent videos.