| 摘要 |
Speech recognition apparatus has a television camera or optical array 6 that views the mouth of a speaker, and a larynogograph 20 that detects movement of the vocal folds. A microphone 5 produces an output in respect of the sound produced by the speaker. After visual processing, signals from the camera 6 are supplied to a pattern matching unit 63 together with signals from a store 64. Signals from the laryngograph are also supplied to the pattern matching unit to resolve ambiguity between speech sounds withsimilar mouth shapes. the output of the microphone is supplied to a noise adaptation unit 52 together with laryngograph output which enables external noise to be identified and rejected. After noise adaptation, the microphone output is subjected to pattern matching with a vocabulary 54 of reference templates. Signals representing the most likely fit after the two pattern matching processes are supplied to a comparator 70 which selects the most likely word spoken, or signals the speaker, via a feedback device 82, to repeat the word spoken. |
