Pattern recognition apparatus for creating multiple systems and combining the multiple systems to improve recognition performance and pattern recognition method

摘要

Provided is a pattern recognition apparatus for creating multiple systems and combining the multiple systems to improve the recognition performance, including a discriminative training unit for constructing model parameters of a second or subsequent system based on an output tendency of a previously-constructed model so as to be different from the output tendency of the previously-constructed model. Accordingly, when multiple systems are combined, the recognition performance can be improved without trials-and-errors.

主权项

1. A pattern recognition apparatus that combines and constructs multiple systems to improve recognition performance, comprising:
a first system having a first model and a first decoder, which outputs a recognition result of the first system, the recognition result of the first system having a first output tendency; a second system having a second model, a second decoder, and a feature transformer, the feature transformer being configured to transform a received feature vector, which is input externally, based on the second model, which has been trained in advance based on the recognition result of the first system, and the feature transformer transforms the feature vector to produce a transformed feature vector that is decoded by the second decoder to output a recognition result of the second system, the recognition result of the second system having a second output tendency; a discriminative training unit that constructs model parameters of the second model in the second system based on the first output tendency of the recognition result of the first system so the second output tendency of the recognition result of the second system is different from the first output tendency of the recognition result of the first system; and a recognizing unit that recognizes a speech pattern or a character pattern in the received feature vector based on the recognition result of the first system and the recognition result of the second system.