TY - GEN
T1 - Domain invariant speech features using a new divergence measure
AU - Wisler, Alan
AU - Berisha, Visar
AU - Liss, Julie
AU - Spanias, Andreas
N1 - Funding Information:
Many thanks to Dr. J. Steller, member of the Polish Academy of Sciences, Institute of Fluid Machinery, and coordinator of the International Cavitation Erosion Test, for sharing with me his deep knowledge on cavitation erosion and providing me with numerous reports on erosion tests.
Publisher Copyright:
© 2014 IEEE.
PY - 2014/4/1
Y1 - 2014/4/1
N2 - Existing speech classification algorithms often perform well when evaluated on training and test data drawn from the same distribution. In practice, however, these distributions are not always the same. In these circumstances, the performance of trained models will likely decrease. In this paper, we discuss an underutilized divergence measure and derive an estimable upper bound on the test error rate that depends on the error rate on the training data and the distance between training and test distributions. Using this bound as motivation, we develop a feature learning algorithm that aims to identify invariant speech features that generalize well to data similar to, but different from, the training set. Comparative results confirm the efficacy of the algorithm on a set of cross-domain speech classification tasks.
AB - Existing speech classification algorithms often perform well when evaluated on training and test data drawn from the same distribution. In practice, however, these distributions are not always the same. In these circumstances, the performance of trained models will likely decrease. In this paper, we discuss an underutilized divergence measure and derive an estimable upper bound on the test error rate that depends on the error rate on the training data and the distance between training and test distributions. Using this bound as motivation, we develop a feature learning algorithm that aims to identify invariant speech features that generalize well to data similar to, but different from, the training set. Comparative results confirm the efficacy of the algorithm on a set of cross-domain speech classification tasks.
KW - Domain adaptation
KW - Feature selection
KW - Machine learning
KW - Pathological speech analysis
UR - http://www.scopus.com/inward/record.url?scp=84946691492&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84946691492&partnerID=8YFLogxK
U2 - 10.1109/SLT.2014.7078553
DO - 10.1109/SLT.2014.7078553
M3 - Conference contribution
AN - SCOPUS:84946691492
T3 - 2014 IEEE Workshop on Spoken Language Technology, SLT 2014 - Proceedings
SP - 77
EP - 82
BT - 2014 IEEE Workshop on Spoken Language Technology, SLT 2014 - Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2014 IEEE Workshop on Spoken Language Technology, SLT 2014
Y2 - 7 December 2014 through 10 December 2014
ER -