Acoustic Space Dimensionality Selection and Combination using the Maximum Entropy Principle

[edit]

Yasser H. Abdel-Haleem
Steve Renals, University of Edinburgh
Neil D. Lawrence, University of Sheffield

in International Conference on Acoustics, Speech and Signal Processing

Related Material

Abstract

In this paper we propose a discriminative approach to acoustic space dimensionality selection based on maximum entropy modelling. We form a set of constraints by composing the acoustic space with the space of phone classes, and use a continuous feature formulation of maximum entropy modelling to select an optimal feature set. The suggested approach has two steps: (1) the selection of the best acoustic space that efficiently and economically represents the acoustic data and its variability; (2) the combination of selected acoustic features in the maximum entropy framework to estimate the posterior probabilities over the phonetic labels given the acoustic input. Specific contributions of this paper include a parameter estimation algorithm (generalized improved iterative scaling) that enables the use of negative features, the parameterization of constraint functions using Gaussian mixture models, and experimental results using the TIMIT database.


@InProceedings{abdelhaleem-acoustic04,
  title = 	 {Acoustic Space Dimensionality Selection and Combination using the Maximum Entropy Principle},
  author = 	 {Yasser H. Abdel-Haleem and Steve Renals and Neil D. Lawrence},
  booktitle = 	 {International Conference on Acoustics, Speech and Signal Processing},
  year = 	 {2004},
  month = 	 {00},
  edit = 	 {https://github.com/lawrennd//publications/edit/gh-pages/_posts/2004-01-01-abdelhaleem-acoustic04.md},
  url =  	 {http://inverseprobability.com/publications/abdelhaleem-acoustic04.html},
  abstract = 	 {In this paper we propose a discriminative approach to acoustic space dimensionality selection based on maximum entropy modelling. We form a set of constraints by composing the acoustic space with the space of phone classes, and use a continuous feature formulation of maximum entropy modelling to select an optimal feature set. The suggested approach has two steps: (1) the selection of the best acoustic space that efficiently and economically represents the acoustic data and its variability; (2) the combination of selected acoustic features in the maximum entropy framework to estimate the posterior probabilities over the phonetic labels given the acoustic input. Specific contributions of this paper include a parameter estimation algorithm (generalized improved iterative scaling) that enables the use of negative features, the parameterization of constraint functions using Gaussian mixture models, and experimental results using the TIMIT database.},
  crossref =  {icassp04},
  key = 	 {AbdelHaleem:acoustic04},
  linkpdf = 	 {ftp://ftp.dcs.shef.ac.uk/home/neil/icassp04.pdf},
  OPTgroup = 	 {}
 

}
%T Acoustic Space Dimensionality Selection and Combination using the Maximum Entropy Principle
%A Yasser H. Abdel-Haleem and Steve Renals and Neil D. Lawrence
%B 
%C International Conference on Acoustics, Speech and Signal Processing
%D 
%F abdelhaleem-acoustic04	
%P --
%R 
%U http://inverseprobability.com/publications/abdelhaleem-acoustic04.html
%X In this paper we propose a discriminative approach to acoustic space dimensionality selection based on maximum entropy modelling. We form a set of constraints by composing the acoustic space with the space of phone classes, and use a continuous feature formulation of maximum entropy modelling to select an optimal feature set. The suggested approach has two steps: (1) the selection of the best acoustic space that efficiently and economically represents the acoustic data and its variability; (2) the combination of selected acoustic features in the maximum entropy framework to estimate the posterior probabilities over the phonetic labels given the acoustic input. Specific contributions of this paper include a parameter estimation algorithm (generalized improved iterative scaling) that enables the use of negative features, the parameterization of constraint functions using Gaussian mixture models, and experimental results using the TIMIT database.
TY  - CPAPER
TI  - Acoustic Space Dimensionality Selection and Combination using the Maximum Entropy Principle
AU  - Yasser H. Abdel-Haleem
AU  - Steve Renals
AU  - Neil D. Lawrence
BT  - International Conference on Acoustics, Speech and Signal Processing
PY  - 2004/01/01
DA  - 2004/01/01	
ID  - abdelhaleem-acoustic04	
SP  - 
EP  - 
L1  - ftp://ftp.dcs.shef.ac.uk/home/neil/icassp04.pdf
UR  - http://inverseprobability.com/publications/abdelhaleem-acoustic04.html
AB  - In this paper we propose a discriminative approach to acoustic space dimensionality selection based on maximum entropy modelling. We form a set of constraints by composing the acoustic space with the space of phone classes, and use a continuous feature formulation of maximum entropy modelling to select an optimal feature set. The suggested approach has two steps: (1) the selection of the best acoustic space that efficiently and economically represents the acoustic data and its variability; (2) the combination of selected acoustic features in the maximum entropy framework to estimate the posterior probabilities over the phonetic labels given the acoustic input. Specific contributions of this paper include a parameter estimation algorithm (generalized improved iterative scaling) that enables the use of negative features, the parameterization of constraint functions using Gaussian mixture models, and experimental results using the TIMIT database.
ER  -

Abdel-Haleem, Y.H., Renals, S. & Lawrence, N.D.. (2004). Acoustic Space Dimensionality Selection and Combination using the Maximum Entropy Principle. International Conference on Acoustics, Speech and Signal Processing :-