Postprocessing of Computation Reduction in Speech Recognition Using Hidden Markov Model

摘要

The present invention is to obtain a higher speech recognition rate by adding a post-processing process to minimize the approximation error, the present invention calculates the contribution of all Gaussian mixtures to the training data in the current leaf despite the high contribution A high voice recognition rate can be obtained by assigning the excluded Gaussian mix to the current riff and excluding the low contribution Gaussian mix from the current riff to minimize the approximation error.