Forward-Decoding Kernel-Based Phone Sequence Recognition

Shantanu Chakrabartty; Gert Cauwenberghs

Forward-Decoding Kernel-Based Phone Sequence Recognition

Shantanu Chakrabartty, Gert Cauwenberghs

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Forward decoding kernel machines (FDKM) combine large-margin classifiers with hidden Markov models (HMM) for maximum a posteriori (MAP) adaptive sequence estimation. State transitions in the sequence are conditioned on observed data using a kernel-based probability model trained with a recursive scheme that deals effectively with noisy and partially labeled data. Training over very large datasets is accomplished using a sparse probabilistic support vector machine (SVM) model based on quadratic entropy, and an on-line stochastic steepest descent algorithm. For speaker-independent continuous phone recognition, FDKM trained over 177,080 samples of the TlMIT database achieves 80.6% recognition accuracy over the full test set, without use of a prior phonetic language model.

Original language	English
Title of host publication	NIPS 2002
Subtitle of host publication	Proceedings of the 15th International Conference on Neural Information Processing Systems
Editors	Suzanna Becker, Sebastian Thrun, Klaus Obermayer
Publisher	MIT Press Journals
Pages	1165-1172
Number of pages	8
ISBN (Electronic)	0262025507, 9780262025508
State	Published - 2002
Event	15th International Conference on Neural Information Processing Systems, NIPS 2002 - Vancouver, Canada Duration: Dec 9 2002 → Dec 14 2002

Publication series

Name	NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems

Conference

Conference	15th International Conference on Neural Information Processing Systems, NIPS 2002
Country/Territory	Canada
City	Vancouver
Period	12/9/02 → 12/14/02

Cite this

Chakrabartty, S., & Cauwenberghs, G. (2002). Forward-Decoding Kernel-Based Phone Sequence Recognition. In S. Becker, S. Thrun, & K. Obermayer (Eds.), NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems (pp. 1165-1172). (NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems). MIT Press Journals.

Chakrabartty, Shantanu ; Cauwenberghs, Gert. / Forward-Decoding Kernel-Based Phone Sequence Recognition. NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems. editor / Suzanna Becker ; Sebastian Thrun ; Klaus Obermayer. MIT Press Journals, 2002. pp. 1165-1172 (NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems).

@inproceedings{52a66c62be734fe1817ac321f2c2f8cf,

title = "Forward-Decoding Kernel-Based Phone Sequence Recognition",

abstract = "Forward decoding kernel machines (FDKM) combine large-margin classifiers with hidden Markov models (HMM) for maximum a posteriori (MAP) adaptive sequence estimation. State transitions in the sequence are conditioned on observed data using a kernel-based probability model trained with a recursive scheme that deals effectively with noisy and partially labeled data. Training over very large datasets is accomplished using a sparse probabilistic support vector machine (SVM) model based on quadratic entropy, and an on-line stochastic steepest descent algorithm. For speaker-independent continuous phone recognition, FDKM trained over 177,080 samples of the TlMIT database achieves 80.6% recognition accuracy over the full test set, without use of a prior phonetic language model.",

author = "Shantanu Chakrabartty and Gert Cauwenberghs",

note = "Publisher Copyright: {\textcopyright} NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems. All rights reserved.; 15th International Conference on Neural Information Processing Systems, NIPS 2002 ; Conference date: 09-12-2002 Through 14-12-2002",

year = "2002",

language = "English",

series = "NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems",

publisher = "MIT Press Journals",

pages = "1165--1172",

editor = "Suzanna Becker and Sebastian Thrun and Klaus Obermayer",

booktitle = "NIPS 2002",

}

Chakrabartty, S & Cauwenberghs, G 2002, Forward-Decoding Kernel-Based Phone Sequence Recognition. in S Becker, S Thrun & K Obermayer (eds), NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems. NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems, MIT Press Journals, pp. 1165-1172, 15th International Conference on Neural Information Processing Systems, NIPS 2002, Vancouver, Canada, 12/9/02.

Forward-Decoding Kernel-Based Phone Sequence Recognition. / Chakrabartty, Shantanu; Cauwenberghs, Gert.
NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems. ed. / Suzanna Becker; Sebastian Thrun; Klaus Obermayer. MIT Press Journals, 2002. p. 1165-1172 (NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Forward-Decoding Kernel-Based Phone Sequence Recognition

AU - Chakrabartty, Shantanu

AU - Cauwenberghs, Gert

PY - 2002

Y1 - 2002

N2 - Forward decoding kernel machines (FDKM) combine large-margin classifiers with hidden Markov models (HMM) for maximum a posteriori (MAP) adaptive sequence estimation. State transitions in the sequence are conditioned on observed data using a kernel-based probability model trained with a recursive scheme that deals effectively with noisy and partially labeled data. Training over very large datasets is accomplished using a sparse probabilistic support vector machine (SVM) model based on quadratic entropy, and an on-line stochastic steepest descent algorithm. For speaker-independent continuous phone recognition, FDKM trained over 177,080 samples of the TlMIT database achieves 80.6% recognition accuracy over the full test set, without use of a prior phonetic language model.

AB - Forward decoding kernel machines (FDKM) combine large-margin classifiers with hidden Markov models (HMM) for maximum a posteriori (MAP) adaptive sequence estimation. State transitions in the sequence are conditioned on observed data using a kernel-based probability model trained with a recursive scheme that deals effectively with noisy and partially labeled data. Training over very large datasets is accomplished using a sparse probabilistic support vector machine (SVM) model based on quadratic entropy, and an on-line stochastic steepest descent algorithm. For speaker-independent continuous phone recognition, FDKM trained over 177,080 samples of the TlMIT database achieves 80.6% recognition accuracy over the full test set, without use of a prior phonetic language model.

UR - http://www.scopus.com/inward/record.url?scp=78049246636&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:78049246636

T3 - NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems

SP - 1165

EP - 1172

BT - NIPS 2002

A2 - Becker, Suzanna

A2 - Thrun, Sebastian

A2 - Obermayer, Klaus

PB - MIT Press Journals

T2 - 15th International Conference on Neural Information Processing Systems, NIPS 2002

Y2 - 9 December 2002 through 14 December 2002

ER -

Chakrabartty S, Cauwenberghs G. Forward-Decoding Kernel-Based Phone Sequence Recognition. In Becker S, Thrun S, Obermayer K, editors, NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems. MIT Press Journals. 2002. p. 1165-1172. (NIPS 2002: Proceedings of the 15th International Conference on Neural Information Processing Systems).

Forward-Decoding Kernel-Based Phone Sequence Recognition

Abstract

Publication series

Conference

Other files and links

Fingerprint

Cite this