Feature selection for log-linear acoustic models

Log-linear acoustic models have been shown to be competitive with Gaussian mixture models in speech recognition. Their high training time can be reduced by feature selection. We compare a simple univariate feature selection algorithm with ReliefF - an efficient multivariate algorithm. An alternative...

Full description

Saved in:
Bibliographic Details
Main Authors: Wiesler, S., Richard, A., Kubo, Y., Schluter, R., Ney, H.
Format: Conference Proceeding
Language:English
Subjects:
acoustic modeling
Acoustics
Complexity theory
feature selection
Hidden Markov models
log-linear models
Optimization
Polynomials
ReliefF
Speech recognition
Training
ℓ 1 -regularization
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Log-linear acoustic models have been shown to be competitive with Gaussian mixture models in speech recognition. Their high training time can be reduced by feature selection. We compare a simple univariate feature selection algorithm with ReliefF - an efficient multivariate algorithm. An alternative to feature selection is ℓ 1 -regularized training, which leads to sparse models. We observe that this gives no speedup when sparse features are used, hence feature selection methods are preferable. For dense features, ℓ 1 -regularization can reduce training and recognition time. We generalize the well known Rprop algorithm for the optimization of ℓ 1 -regularized functions. Experiments on the Wall Street Journal corpus showed that a large number of sparse features could be discarded without loss of performance. A strong regularization led to slight performance degradations, but can be useful on large tasks, where training the full model is not tractable.
ISSN:1520-6149
2379-190X
DOI:10.1109/ICASSP.2011.5947560