Abstract
Our work is concerned with the problem of classifying a sample as belonging or not to a Target Class (TC) by means of a discriminant classifier. To yield this goal, the classifier must be trained with samples of the TC and Non-Target Class (NTC). The problem arises when, as in many real applications, the number of the NTC examples is much higher than that of the TC, leading the classifier, if all the examples are used in training, to overlearn the NTC. This paper is focused in the task of extracting, from the pool of NTC representatives, the most discriminant training subset with regard to the TC, and with the optimum size: too few NTC examples would be non-representative and the classifier tends to missaccept, and with too many the classifier tends to misreject. A new heuristic search method is presented that improves the performance of the traditional problem solution, the random selection, eliminating the random behavior of the system. To prove this advantages the new technique will be applied to speaker verification task by means of neural networks, achieving in this way a more competitive system.
This work has been supported by Ministerio de Ciencia y Tecnología, Spain, under Project TIC2000-1669-C03
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsPreview
Unable to display preview. Download preview PDF.
References
Auckenthaler, R., Carey, M., Harvey, L.: Score Normalization for Text-Independent Speaker Verification Systems. Digital Signal Processing 10, 42–54 (2000)
Martin, A., Przyboki, M.: The NIST Speaker Recognition Evaluations: 1996-2001. In: Proc. of the 2001: The Speaker Odyssey, The Speaker Recognition Workshop (2001)
Ortega-Garcia, J., Gonzalez-Rodriguez, J., Marrero-Aguiar, V.: An Approach to Forensic Speaker Verification Using AHUMADA Large Speech Corpus in Spanish. Speech Communication 31, 255–264 (2000)
Vivaracho, C., Ortega-Garcia, J., Alonso, L., Moro, Q.: A Comparative Study of MLP-based Artificial Neural Networks in Text-Independent Speaker Verification against GMM-based Systems. In: Proc. of Eurospeech 2001, vol. 3, pp. 1753–1756 (2001)
Yegnanarayana, B., Kishore, S.: AANN: an Alternative to GMM for Pattern Recognition. Neural Networks 15(3), 459–469 (2002)
Vivaracho, C., Ortega-Garcia, J., Alonso, L., Moro, Q.: Improving the Competitiveness of Discriminant Neural Networks in Speaker Verification. In: Proc. Eurospeech 2003 (2003) (to appear)
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2003 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Vivaracho, C.E., Ortega-Garcia, J., Alonso, L., Moro, Q.I. (2003). Extracting the Most Discriminant Subset from a Pool of Candidates to Optimize Discriminant Classifier Training. In: Zhong, N., Raś, Z.W., Tsumoto, S., Suzuki, E. (eds) Foundations of Intelligent Systems. ISMIS 2003. Lecture Notes in Computer Science(), vol 2871. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-39592-8_92
Download citation
DOI: https://doi.org/10.1007/978-3-540-39592-8_92
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-20256-1
Online ISBN: 978-3-540-39592-8
eBook Packages: Springer Book Archive