Skip to content

Detail of publication


Jan Trmal : Spatio-temporal structure of feature vectors in neural network adaptation . 2012.

Download PDF



This doctoral thesis aims at research in the field of neural networks adaptation and in the field of speaker adaptive training, with special attention to the application of both in the field of automatic speech recognition. Both these technologies, i.e. adaptation and speaker adaptive training are often used in the area of speech recognition in the context of GMM/HMM modeling framework. In that context, they pose one possible approach to improving recognition accuracy, often at a cost of an insignificant increase of computational complexity. The crucial assumptions of both these techniques, i.e. of the speaker adaptation and of the speaker adaptive training, are realistic and can be ensured relatively easily. Therefore, it is desirable to have similar techniques developed even for hybrid (i.e. non-GMM/HMM) speech recognition systems. The goal of this thesis was to develop such a method and to experimentally evaluate its influence on the accuracy of the speech recognition system.

Detail of publication

Title: Spatio-temporal structure of feature vectors in neural network adaptation
Author: Jan Trmal
Language: Czech
Date of publication: 15 Mar 2012
Year: 2012
Type of publication: Habilitation and dissertation theses
/ 2012-03-15 11:23:40 /


acoustic modeling, neural networks, adaptation, speech recognition, speaker adaptive training


 author = {Jan Trmal},
 title = {Spatio-temporal structure of feature vectors in neural network adaptation},
 year = {2012},
 url = {},