Automatic lipreading to enhance speech recognition (speech reading)

摘要: Automatic recognition of the acoustic speech signal alone is inaccurate and computationally expensive. Additional sources information, such as lipreading (or speechreading), should enhance automatic recognition, just used by humans to when degraded. This paper describes an system which has been developed. A commercial device performs independently system. The domain restricted isolated utterances speaker dependent recognition. The faces a solid state camera sends digitized video minicomputer with custom processing hardware. data sampled during utterance then reduced template consisting visual parameter time sequences. distances between incoming all trained templates for each in vocabulary are computed candidate obtained. combination candidates shown yield final accuracy greatly exceeds alone. Practical considerations possible enhancement independent continuous systems also discussed.