Publication detail

An Efficient training algorithm for frame-based phoneme classification

PFEIFER, V. VRBA, K. MÜLLER, J.

Original Title

An Efficient training algorithm for frame-based phoneme classification

English Title

An Efficient training algorithm for frame-based phoneme classification

Type

conference paper

Language

en

Original Abstract

This paper proposed new afficient training algorithm for a hiearchical frame-based phoneme classifier. THis hiearchical structure represents a simple metric over the tree root structure. The proposed classification function is based on the linear classification problem where each phoneme has its own weight vector which is trained during the learning phase. The results clearly shows that our proposed algorithm outperforms all the others. All the evaluation have been performed over the timit speech corpus.

English abstract

This paper proposed new afficient training algorithm for a hiearchical frame-based phoneme classifier. THis hiearchical structure represents a simple metric over the tree root structure. The proposed classification function is based on the linear classification problem where each phoneme has its own weight vector which is trained during the learning phase. The results clearly shows that our proposed algorithm outperforms all the others. All the evaluation have been performed over the timit speech corpus.

Keywords

classifier, features, kernel, optimalization, phoneme, prototypes, speech, tree

RIV year

2010

Released

07.09.2010

ISBN

978-80-248-2261-7

Book

Research in Telecomunication Technology - RTT 2010

Pages from

1

Pages to

5

Pages count

5

BibTex


@inproceedings{BUT35247,
  author="Václav {Pfeifer} and Kamil {Vrba} and Jakub {Müller}",
  title="An Efficient training algorithm for frame-based phoneme classification",
  annote="This paper proposed new afficient training algorithm for a hiearchical frame-based phoneme classifier. THis hiearchical structure represents a simple metric over the tree root structure. The proposed classification function is based on the linear classification problem where each phoneme has its own weight vector which is trained during the learning phase. The results clearly shows that our proposed algorithm outperforms all the others. All the evaluation have been performed over the timit speech corpus.",
  booktitle="Research in Telecomunication Technology - RTT 2010",
  chapter="35247",
  howpublished="electronic, physical medium",
  year="2010",
  month="september",
  pages="1--5",
  type="conference paper"
}