Large vocabulary speech recognition using neural prediction model

Abstract

The authors present improvements in the neural prediction model. The improvements include the introduction of backward prediction in the pattern predictors and the modification of the prediction error measure with covariance matrices. Using the demisyllable as a subword recognition unit, speaker-dependent large vocabulary recognition experiments were carried out. Results indicate a 97.6% recognition accuracy for a 5000-word test set, and the effectiveness of the proposed model improvements and the demisyllable subword units was confirmed.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>