Cluster-Based Adaptation Using Density Forest for HMM Phone Recognition

Publikation: Bidrag til bog/antologi/rapport/konference proceedingKonferenceartikel i proceedingForskningpeer review

127 Downloads (Pure)

Resumé

The dissimilarity between the training and test data in speech recognition systems is known to have a considerable effect on the recognition accuracy. To solve this problem, we use density forest to cluster the data and use maximum a posteriori (MAP) method to build a cluster-based adapted Gaussian mixture models (GMMs) in HMM speech recognition. Specifically, a set of bagged versions of the training data for each state in the HMM is generated, and each of these versions is used to generate one GMM and one tree in the density forest. Thereafter, an acoustic model forest is built by replacing the data of each leaf (cluster) in each tree with the corresponding GMM adapted by the leaf data using the MAP method. The results show that the proposed approach achieves 3:8% (absolute) lower phone error rate compared with the standard HMM/GMM and 0:8% (absolute) lower PER compared with bagged HMM/GMM.
OriginalsprogEngelsk
TitelSignal Processing Conference (EUSIPCO), 2014 Proceedings of the 22nd European
ForlagIEEE Press
Publikationsdato2014
Sider2065-2069
ISBN (Trykt)978-0-9928626-1-9
StatusUdgivet - 2014
Begivenhed22nd European Signal Processing Conference - Lisbon, Portugal
Varighed: 1 sep. 20145 sep. 2014

Konference

Konference22nd European Signal Processing Conference
LandPortugal
ByLisbon
Periode01/09/201405/09/2014
NavnProceedings of the European Signal Processing Conference
ISSN2076-1465

Fingerprint

Speech recognition
Acoustics

Citer dette

Abou-Zleikha, M., Tan, Z-H., Christensen, M. G., & Jensen, S. H. (2014). Cluster-Based Adaptation Using Density Forest for HMM Phone Recognition. I Signal Processing Conference (EUSIPCO), 2014 Proceedings of the 22nd European (s. 2065-2069). IEEE Press. Proceedings of the European Signal Processing Conference
Abou-Zleikha, Mohamed ; Tan, Zheng-Hua ; Christensen, Mads Græsbøll ; Jensen, Søren Holdt. / Cluster-Based Adaptation Using Density Forest for HMM Phone Recognition. Signal Processing Conference (EUSIPCO), 2014 Proceedings of the 22nd European. IEEE Press, 2014. s. 2065-2069 (Proceedings of the European Signal Processing Conference).
@inproceedings{e0dc0e6a3f7443beafbce2632e377052,
title = "Cluster-Based Adaptation Using Density Forest for HMM Phone Recognition",
abstract = "The dissimilarity between the training and test data in speech recognition systems is known to have a considerable effect on the recognition accuracy. To solve this problem, we use density forest to cluster the data and use maximum a posteriori (MAP) method to build a cluster-based adapted Gaussian mixture models (GMMs) in HMM speech recognition. Specifically, a set of bagged versions of the training data for each state in the HMM is generated, and each of these versions is used to generate one GMM and one tree in the density forest. Thereafter, an acoustic model forest is built by replacing the data of each leaf (cluster) in each tree with the corresponding GMM adapted by the leaf data using the MAP method. The results show that the proposed approach achieves 3:8{\%} (absolute) lower phone error rate compared with the standard HMM/GMM and 0:8{\%} (absolute) lower PER compared with bagged HMM/GMM.",
author = "Mohamed Abou-Zleikha and Zheng-Hua Tan and Christensen, {Mads Gr{\ae}sb{\o}ll} and Jensen, {S{\o}ren Holdt}",
year = "2014",
language = "English",
isbn = "978-0-9928626-1-9",
series = "Proceedings of the European Signal Processing Conference",
publisher = "IEEE Press",
pages = "2065--2069",
booktitle = "Signal Processing Conference (EUSIPCO), 2014 Proceedings of the 22nd European",

}

Abou-Zleikha, M, Tan, Z-H, Christensen, MG & Jensen, SH 2014, Cluster-Based Adaptation Using Density Forest for HMM Phone Recognition. i Signal Processing Conference (EUSIPCO), 2014 Proceedings of the 22nd European. IEEE Press, Proceedings of the European Signal Processing Conference, s. 2065-2069, 22nd European Signal Processing Conference, Lisbon, Portugal, 01/09/2014.

Cluster-Based Adaptation Using Density Forest for HMM Phone Recognition. / Abou-Zleikha, Mohamed; Tan, Zheng-Hua; Christensen, Mads Græsbøll; Jensen, Søren Holdt.

Signal Processing Conference (EUSIPCO), 2014 Proceedings of the 22nd European. IEEE Press, 2014. s. 2065-2069 (Proceedings of the European Signal Processing Conference).

Publikation: Bidrag til bog/antologi/rapport/konference proceedingKonferenceartikel i proceedingForskningpeer review

TY - GEN

T1 - Cluster-Based Adaptation Using Density Forest for HMM Phone Recognition

AU - Abou-Zleikha, Mohamed

AU - Tan, Zheng-Hua

AU - Christensen, Mads Græsbøll

AU - Jensen, Søren Holdt

PY - 2014

Y1 - 2014

N2 - The dissimilarity between the training and test data in speech recognition systems is known to have a considerable effect on the recognition accuracy. To solve this problem, we use density forest to cluster the data and use maximum a posteriori (MAP) method to build a cluster-based adapted Gaussian mixture models (GMMs) in HMM speech recognition. Specifically, a set of bagged versions of the training data for each state in the HMM is generated, and each of these versions is used to generate one GMM and one tree in the density forest. Thereafter, an acoustic model forest is built by replacing the data of each leaf (cluster) in each tree with the corresponding GMM adapted by the leaf data using the MAP method. The results show that the proposed approach achieves 3:8% (absolute) lower phone error rate compared with the standard HMM/GMM and 0:8% (absolute) lower PER compared with bagged HMM/GMM.

AB - The dissimilarity between the training and test data in speech recognition systems is known to have a considerable effect on the recognition accuracy. To solve this problem, we use density forest to cluster the data and use maximum a posteriori (MAP) method to build a cluster-based adapted Gaussian mixture models (GMMs) in HMM speech recognition. Specifically, a set of bagged versions of the training data for each state in the HMM is generated, and each of these versions is used to generate one GMM and one tree in the density forest. Thereafter, an acoustic model forest is built by replacing the data of each leaf (cluster) in each tree with the corresponding GMM adapted by the leaf data using the MAP method. The results show that the proposed approach achieves 3:8% (absolute) lower phone error rate compared with the standard HMM/GMM and 0:8% (absolute) lower PER compared with bagged HMM/GMM.

M3 - Article in proceeding

SN - 978-0-9928626-1-9

T3 - Proceedings of the European Signal Processing Conference

SP - 2065

EP - 2069

BT - Signal Processing Conference (EUSIPCO), 2014 Proceedings of the 22nd European

PB - IEEE Press

ER -

Abou-Zleikha M, Tan Z-H, Christensen MG, Jensen SH. Cluster-Based Adaptation Using Density Forest for HMM Phone Recognition. I Signal Processing Conference (EUSIPCO), 2014 Proceedings of the 22nd European. IEEE Press. 2014. s. 2065-2069. (Proceedings of the European Signal Processing Conference).