A review of audio-visual speech recognition

Speech is the most important tool of interaction among human beings. This has inspired researchers to study further on speech recognition and develop a computer system that is able to integrate and understand human speech. But acoustic noisy environment can highly contaminate audio speech and affect the overall recognition performance. Thus, Audio-Visual Speech Recognition (AVSR) is designed to overcome the problems by utilising visual images which are unaffected by noise. The aim of this paper is to discuss the AVSR structures, which includes the front end processes, audio-visual data corpus used, recent works and accuracy estimation methods