This shows you the differences between two versions of the page.

Link to this comparison view

language_modelling_using_som_network [2014/03/27 21:34] (current)
bziolko created
Line 1: Line 1:
 +Leszek Gajecki (Poland)
 +**Language modelling using SOM network**
 +University of Information Technologuy and Management, Rzeszów
 +The Language Model module is a part of the Large Vocabulary Continuous Speech Recognition System (LVCSR). Ngram model and its derivates are typically applied solutions. They are simple and computationally fast solutions. Some improvements of this models help in data sparseness problem. This solution is well suited for languages with strict words’ order (for example English), while other language require more training examples. Proposed solution is a model based on Kohonen’s Self Organised Maps, which smooth probabilities, but also can be used for languages with less strict word’s order (like Polish). Presented results show improvement in comparison to trigram and bigram models in dictation task.
Copyright © XXII PVC Organizing Committee 2013. All Rights Reserved.