Feature adaptation for robust mobile speech recognition

Hyeopwoo Lee; Dongsuk Yook

doi:10.1109/TCE.2012.6415011

Feature adaptation for robust mobile speech recognition

Research output: Contribution to journal › Article › peer-review

Abstract

Feature adaptation such as feature space maximum likelihood linear regression (FMLLR) is useful for robust mobile speech recognition. However, as the amount of adaptation data increases, feature adaptation performance becomes saturated quickly due to its limitation of global transformation. To handle this problem, we propose regression tree based FMLLR which can adopt multiple transformations as the amount of adaptation data increases. An experimental result shows that the proposed method reduces the recognition error by 11.8% further for speaker adaptation task and by 13.6% further for noisy environment adaptation task compared to the conventional method.

Original language	English
Article number	6415011
Pages (from-to)	1393-1398
Number of pages	6
Journal	IEEE Transactions on Consumer Electronics
Volume	58
Issue number	4
DOIs	https://doi.org/10.1109/TCE.2012.6415011
Publication status	Published - 2012

Bibliographical note

Funding Information:
1This work was supported by the Korea Research Foundation (KRF) grant funded by the Korea government (MEST) (No. 2011-0002906).

Keywords

Speech recognition
environment adaptation
feature adaptation
feature spacemaximum likelihood linear regression (FMLLR)
regression tree
speaker adaptation

ASJC Scopus subject areas

Media Technology
Electrical and Electronic Engineering

Access to Document

10.1109/TCE.2012.6415011

Cite this

@article{4f2e195938dd462ca13c03a399db2cdc,

title = "Feature adaptation for robust mobile speech recognition",

abstract = "Feature adaptation such as feature space maximum likelihood linear regression (FMLLR) is useful for robust mobile speech recognition. However, as the amount of adaptation data increases, feature adaptation performance becomes saturated quickly due to its limitation of global transformation. To handle this problem, we propose regression tree based FMLLR which can adopt multiple transformations as the amount of adaptation data increases. An experimental result shows that the proposed method reduces the recognition error by 11.8% further for speaker adaptation task and by 13.6% further for noisy environment adaptation task compared to the conventional method.",

keywords = "Speech recognition, environment adaptation, feature adaptation, feature spacemaximum likelihood linear regression (FMLLR), regression tree, speaker adaptation",

author = "Hyeopwoo Lee and Dongsuk Yook",

note = "Funding Information: 1This work was supported by the Korea Research Foundation (KRF) grant funded by the Korea government (MEST) (No. 2011-0002906).",

year = "2012",

doi = "10.1109/TCE.2012.6415011",

language = "English",

volume = "58",

pages = "1393--1398",

journal = "IEEE Transactions on Consumer Electronics",

issn = "0098-3063",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

number = "4",

}

TY - JOUR

T1 - Feature adaptation for robust mobile speech recognition

AU - Lee, Hyeopwoo

AU - Yook, Dongsuk

N1 - Funding Information: 1This work was supported by the Korea Research Foundation (KRF) grant funded by the Korea government (MEST) (No. 2011-0002906).

PY - 2012

Y1 - 2012

N2 - Feature adaptation such as feature space maximum likelihood linear regression (FMLLR) is useful for robust mobile speech recognition. However, as the amount of adaptation data increases, feature adaptation performance becomes saturated quickly due to its limitation of global transformation. To handle this problem, we propose regression tree based FMLLR which can adopt multiple transformations as the amount of adaptation data increases. An experimental result shows that the proposed method reduces the recognition error by 11.8% further for speaker adaptation task and by 13.6% further for noisy environment adaptation task compared to the conventional method.

AB - Feature adaptation such as feature space maximum likelihood linear regression (FMLLR) is useful for robust mobile speech recognition. However, as the amount of adaptation data increases, feature adaptation performance becomes saturated quickly due to its limitation of global transformation. To handle this problem, we propose regression tree based FMLLR which can adopt multiple transformations as the amount of adaptation data increases. An experimental result shows that the proposed method reduces the recognition error by 11.8% further for speaker adaptation task and by 13.6% further for noisy environment adaptation task compared to the conventional method.

KW - Speech recognition

KW - environment adaptation

KW - feature adaptation

KW - feature spacemaximum likelihood linear regression (FMLLR)

KW - regression tree

KW - speaker adaptation

UR - http://www.scopus.com/inward/record.url?scp=84873806083&partnerID=8YFLogxK

U2 - 10.1109/TCE.2012.6415011

DO - 10.1109/TCE.2012.6415011

M3 - Article

AN - SCOPUS:84873806083

SN - 0098-3063

VL - 58

SP - 1393

EP - 1398

JO - IEEE Transactions on Consumer Electronics

JF - IEEE Transactions on Consumer Electronics

IS - 4

M1 - 6415011

ER -

Feature adaptation for robust mobile speech recognition

Abstract

Bibliographical note

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this