PURPOSE: To improve performance in a conversation recognition system using only voice or visual oral position information under an environment where a particularly disadvantage noise is rich.
CONSTITUTION: This system is a conversation recognition system for recognizing speaking belonging to a prescribed set consisting of an allowable speaking candidate, and is constituted of a voice feature extracting device 24 for converting a signal showing voice conversation to a set of the signal having a corresponding voice feature vector, a motional visual feature extracting device 14 for converting an attending signal showing a motional face feature accompanying voice conversation generation to the set of the signal having a corresponding visual feature vector and a neural network classification device 200 for generating a conditional probability distribution in the speaking candidate of allowable conversation by receiving and operating the set in the motional voice feature and the visual feature vectors provided respectively by the voice and the motional visual feature extracting devices 24, 14.
GUREGORII JIEI UORUFU
AARU AI REBIN