ITRW on Non-Linear Speech Processing (NOLISP 07) Paris, France May 22-25, 2007

Hybrid Models for Automatic Speech Recognition: A Comparison of Classical ANN and Kernel-Based Methods

Ana I. García-Moral, Rubén Solera-Urena, Carmen Peláez-Moreno, Fernando Díaz-de-María

Department of Signal Theory and Communicationos, University Carlos III Madrid, Leganés (Madrid), Spain

Support Vector Machines (SVM) are state-of-the-art methods for machine learning but share with more classical Artificial Neural Networks (ANN) the difficulty of their application to temporally variable input patterns. This is the case in Automatic Speech Recognition (ASR). In this paper we have recalled the solutions provided in the past for ANN and applied them to SVMs performing a comparison between them. Preliminary results show a similar behaviour which results encouraging if we take into account the novelty of the SVM systems in comparison with classical ANNs. The envisioned ways of improvement are outlined in the paper.

Full Paper

Bibliographic reference.  García-Moral, Ana I. / Solera-Urena, Rubén / Peláez-Moreno, Carmen / Díaz-de-María, Fernando (2007): "Hybrid models for automatic speech recognition: a comparison of classical ANN and kernel-based methods", In NOLISP-2007, 51-54.