Variable Selection Linear Regression for Robust Speech Recognition

Yu TSAO; Ting-Yao HU; Sakriani SAKTI; Satoshi NAKAMURA; Lin-shan LEE

doi:10.1587/transinf.E97.D.1477

Variable Selection Linear Regression for Robust Speech Recognition

Yu TSAO, Ting-Yao HU, Sakriani SAKTI, Satoshi NAKAMURA, Lin-shan LEE

Full Text Views

0

Cite this

Summary :

This study proposes a variable selection linear regression (VSLR) adaptation framework to improve the accuracy of automatic speech recognition (ASR) with only limited and unlabeled adaptation data. The proposed framework can be divided into three phases. The first phase prepares multiple variable subsets by applying a ranking filter to the original regression variable set. The second phase determines the best variable subset based on a pre-determined performance evaluation criterion and computes a linear regression (LR) mapping function based on the determined subset. The third phase performs adaptation in either model or feature spaces. The three phases can select the optimal components and remove redundancies in the LR mapping function effectively and thus enable VSLR to provide satisfactory adaptation performance even with a very limited number of adaptation statistics. We formulate model space VSLR and feature space VSLR by integrating the VS techniques into the conventional LR adaptation systems. Experimental results on the Aurora-4 task show that model space VSLR and feature space VSLR, respectively, outperform standard maximum likelihood linear regression (MLLR) and feature space MLLR (fMLLR) and their extensions, with notable word error rate (WER) reductions in a per-utterance unsupervised adaptation manner.

Publication: IEICE TRANSACTIONS on Information Vol.E97-D No.6 pp.1477-1487

Publication Date: 2014/06/01

Publicized

Online ISSN: 1745-1361

DOI: 10.1587/transinf.E97.D.1477

Type of Manuscript: Special Section PAPER (Special Section on Advances in Modeling for Real-world Speech Information Processing and its Application)

Category: Speech Recognition

Authors

Yu TSAO
  Academia Sinica
Ting-Yao HU
  National Taiwan University
Sakriani SAKTI
  Nara Institute of Science and Technology
Satoshi NAKAMURA
  Nara Institute of Science and Technology
Lin-shan LEE
  National Taiwan University

Keyword

variable selection, linear regression, MLLR, fMLLR, model space adaptation, feature space adaptation

Cite this

Copy

Yu TSAO, Ting-Yao HU, Sakriani SAKTI, Satoshi NAKAMURA, Lin-shan LEE, "Variable Selection Linear Regression for Robust Speech Recognition" in IEICE TRANSACTIONS on Information, vol. E97-D, no. 6, pp. 1477-1487, June 2014, doi: 10.1587/transinf.E97.D.1477.
Abstract: This study proposes a variable selection linear regression (VSLR) adaptation framework to improve the accuracy of automatic speech recognition (ASR) with only limited and unlabeled adaptation data. The proposed framework can be divided into three phases. The first phase prepares multiple variable subsets by applying a ranking filter to the original regression variable set. The second phase determines the best variable subset based on a pre-determined performance evaluation criterion and computes a linear regression (LR) mapping function based on the determined subset. The third phase performs adaptation in either model or feature spaces. The three phases can select the optimal components and remove redundancies in the LR mapping function effectively and thus enable VSLR to provide satisfactory adaptation performance even with a very limited number of adaptation statistics. We formulate model space VSLR and feature space VSLR by integrating the VS techniques into the conventional LR adaptation systems. Experimental results on the Aurora-4 task show that model space VSLR and feature space VSLR, respectively, outperform standard maximum likelihood linear regression (MLLR) and feature space MLLR (fMLLR) and their extensions, with notable word error rate (WER) reductions in a per-utterance unsupervised adaptation manner.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.E97.D.1477/_p

Copy

@ARTICLE{e97-d_6_1477,
author={Yu TSAO, Ting-Yao HU, Sakriani SAKTI, Satoshi NAKAMURA, Lin-shan LEE, },
journal={IEICE TRANSACTIONS on Information},
title={Variable Selection Linear Regression for Robust Speech Recognition},
year={2014},
volume={E97-D},
number={6},
pages={1477-1487},
abstract={This study proposes a variable selection linear regression (VSLR) adaptation framework to improve the accuracy of automatic speech recognition (ASR) with only limited and unlabeled adaptation data. The proposed framework can be divided into three phases. The first phase prepares multiple variable subsets by applying a ranking filter to the original regression variable set. The second phase determines the best variable subset based on a pre-determined performance evaluation criterion and computes a linear regression (LR) mapping function based on the determined subset. The third phase performs adaptation in either model or feature spaces. The three phases can select the optimal components and remove redundancies in the LR mapping function effectively and thus enable VSLR to provide satisfactory adaptation performance even with a very limited number of adaptation statistics. We formulate model space VSLR and feature space VSLR by integrating the VS techniques into the conventional LR adaptation systems. Experimental results on the Aurora-4 task show that model space VSLR and feature space VSLR, respectively, outperform standard maximum likelihood linear regression (MLLR) and feature space MLLR (fMLLR) and their extensions, with notable word error rate (WER) reductions in a per-utterance unsupervised adaptation manner.},
keywords={},
doi={10.1587/transinf.E97.D.1477},
ISSN={1745-1361},
month={June},}

Copy

TY - JOUR
TI - Variable Selection Linear Regression for Robust Speech Recognition
T2 - IEICE TRANSACTIONS on Information
SP - 1477
EP - 1487
AU - Yu TSAO
AU - Ting-Yao HU
AU - Sakriani SAKTI
AU - Satoshi NAKAMURA
AU - Lin-shan LEE
PY - 2014
DO - 10.1587/transinf.E97.D.1477
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E97-D
IS - 6
JA - IEICE TRANSACTIONS on Information
Y1 - June 2014
AB - This study proposes a variable selection linear regression (VSLR) adaptation framework to improve the accuracy of automatic speech recognition (ASR) with only limited and unlabeled adaptation data. The proposed framework can be divided into three phases. The first phase prepares multiple variable subsets by applying a ranking filter to the original regression variable set. The second phase determines the best variable subset based on a pre-determined performance evaluation criterion and computes a linear regression (LR) mapping function based on the determined subset. The third phase performs adaptation in either model or feature spaces. The three phases can select the optimal components and remove redundancies in the LR mapping function effectively and thus enable VSLR to provide satisfactory adaptation performance even with a very limited number of adaptation statistics. We formulate model space VSLR and feature space VSLR by integrating the VS techniques into the conventional LR adaptation systems. Experimental results on the Aurora-4 task show that model space VSLR and feature space VSLR, respectively, outperform standard maximum likelihood linear regression (MLLR) and feature space MLLR (fMLLR) and their extensions, with notable word error rate (WER) reductions in a per-utterance unsupervised adaptation manner.
ER -