Görsel–işitsel konuşma tanıma’da veri kaynaştırma teknikleri (Information fusion techniques in audio-visual speech recognition)

摘要

It is well known that human perception of speech relies both on audio and visual information. However, the physiology of information fusion process in humans is still indefinite which attracts scientists' attention to information fusion process for audio-visual speech recognition. In this work, a novel tandem hybrid approach is introduced for an efficient audio-visual speech recognition system and the performance of the proposed technique is experimentally compared with the widely used Multiple Stream Hidden Markov Model (MSHMM) approach.