Acoustic and visual input speech recognition system - monitors lip and mouth movements by video camera to provide motion vector input to neural network based speech identification unit.

摘要

The speech identification system has both acoustic data input from a microphone (20) and video information obtained by monitoring lips and mouth movements of speaker with a video camera (10). A position vector generator (14) provides input to an interpolator (26) that generates a characteristic for input to a speech classification unit (200). This also receives the output of a spectral analyser operating on the acoustic signals. The speech classification unit is in the form of a multi layer time delay neural network. This responds to the spectral data for successful certification. USE/ADVANTAGE - Improves probability of correct speech identification by persons with different accents, different sex, different speed of speech, different degrees of coherency.