Comparing Linear Feature Space Transformations for Correlated Features

Authors:
Daniel Vásquez;Rainer Gruhn;Raymond Brueckner;Wolfgang Minker
Affiliations:
Department of Information Technology, University of Ulm, Ulm, Germany and Harman/Becker Automotive Systems, Speech Dialog Systems, Ulm, Germany;Department of Information Technology, University of Ulm, Ulm, Germany and Harman/Becker Automotive Systems, Speech Dialog Systems, Ulm, Germany;Harman/Becker Automotive Systems, Speech Dialog Systems, Ulm, Germany;Department of Information Technology, University of Ulm, Ulm, Germany
Venue:
PIT '08 Proceedings of the 4th IEEE tutorial and research workshop on Perception and Interactive Technologies for Speech-Based Systems: Perception in Multimodal Dialogue Systems
Year:
2008

Citing 3
Cited 0

Eigenfaces vs. Fisherfaces: Recognition Using Class Specific Linear Projection

IEEE Transactions on Pattern Analysis and Machine Intelligence
Spoken Language Processing: A Guide to Theory, Algorithm, and System Development

Spoken Language Processing: A Guide to Theory, Algorithm, and System Development
Investigation of silicon auditory models and generalization of linear discriminant analysis for improved speech recognition

Investigation of silicon auditory models and generalization of linear discriminant analysis for improved speech recognition

Quantified Score

Hi-index	0.00

Visualization

Abstract

In automatic speech recognition, a common method to decorrelate features and to reduce feature space dimensionality is Linear Discriminant Analysis (LDA). In this paper, the performance of LDA has been compared with other linear feature space transformation schemes, as many alternative methods have been suggested and lead to higher recognition accuracy in some cases. Different approaches such as MLLT, HLDA, SHLDA, PCA, and combined schemes were implemented and compared. Experiments show that all methods lead to similar results.In addition, recent research has shown that the LDA algorithm is unreliable if the input features of LDA are strongly correlated. In this paper a stable solution to the correlated feature problem, consisting of a concatenation scheme with PCA and LDA, is proposed and verified. Finally, several transformation algorithms are evaluated on uncorrelated and strongly correlated features.