Knowledge-based quadratic discriminant analysis for phonetic classification

Heyun Huang; Yang Liu; Louis Ten Bosch; Bert Cranen; Lou Boves

doi:10.1109/ICASSP.2012.6288831

Knowledge-based quadratic discriminant analysis for phonetic classification

Heyun Huang^*
, Yang Liu
, Louis Ten Bosch
, Bert Cranen
, Lou Boves

^*Corresponding author for this work

Department of Computer Science

Research output: Chapter in book/report/conference proceeding › Conference proceeding › peer-review

4 Citations (Scopus)

Abstract

Modeling the second-order statistics of articulatory trajectories is likely to improve the performance in classifying phone segments compared to using only linear combinations of MFCCs. Nevertheless, the extremely high dimensionality of the feature space spanned by a combination of monomials of degree-1 and degree-2 makes it difficult to effectively exploit the discriminative information in the full covariance matrix. This paper proposes a novel algorithm, dubbed Knowledge-based Quadratic Discriminant Analysis (KnQDA), for reducing the number of dimensions of the space spanned by degree-1 and degree-2 monomials by using phonetic knowledge for selecting the set of degree-2 monomials that are most likely to improve classification. KnQDA seeks a trade-off between overfitting and undertraining, which further improves the learnability. Binary classifications on all pairs of phones in TIMIT show the effectiveness of the proposed method, especially on those phone pairs that overlap strongly in the linear feature space.

Original language	English
Title of host publication	2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings
Publisher	IEEE
Pages	4145-4148
Number of pages	4
ISBN (Electronic)	9781467300445
ISBN (Print)	9781467300469
DOIs	https://doi.org/10.1109/ICASSP.2012.6288831
Publication status	Published - 25 Mar 2012
Event	2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Kyoto, Japan Duration: 25 Mar 2012 → 30 Mar 2012 https://doi.org/10.1109/ICASSP15465.2012 (Conference proceeding)

Publication series

Name	ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings
ISSN (Print)	1520-6149

Conference

Conference	2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012
Abbreviated title	ICASSP 2012
Country/Territory	Japan
City	Kyoto
Period	25/03/12 → 30/03/12
Internet address	https://doi.org/10.1109/ICASSP15465.2012 (Conference proceeding)

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 9 Industry, Innovation, and Infrastructure

User-Defined Keywords

Dimensionality Reduction
Knowledge-Based Quadratic Discriminant Analysis
Phone Classification
TIMIT

Access to Document

10.1109/ICASSP.2012.6288831

Cite this

Huang, H., Liu, Y., Ten Bosch, L., Cranen, B., & Boves, L. (2012). Knowledge-based quadratic discriminant analysis for phonetic classification. In 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings (pp. 4145-4148). (ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings). IEEE. https://doi.org/10.1109/ICASSP.2012.6288831

@inproceedings{02e3563b6d11463c8f8deda081a14be5,

title = "Knowledge-based quadratic discriminant analysis for phonetic classification",

abstract = "Modeling the second-order statistics of articulatory trajectories is likely to improve the performance in classifying phone segments compared to using only linear combinations of MFCCs. Nevertheless, the extremely high dimensionality of the feature space spanned by a combination of monomials of degree-1 and degree-2 makes it difficult to effectively exploit the discriminative information in the full covariance matrix. This paper proposes a novel algorithm, dubbed Knowledge-based Quadratic Discriminant Analysis (KnQDA), for reducing the number of dimensions of the space spanned by degree-1 and degree-2 monomials by using phonetic knowledge for selecting the set of degree-2 monomials that are most likely to improve classification. KnQDA seeks a trade-off between overfitting and undertraining, which further improves the learnability. Binary classifications on all pairs of phones in TIMIT show the effectiveness of the proposed method, especially on those phone pairs that overlap strongly in the linear feature space.",

keywords = "Dimensionality Reduction, Knowledge-Based Quadratic Discriminant Analysis, Phone Classification, TIMIT",

author = "Heyun Huang and Yang Liu and \{Ten Bosch\}, Louis and Bert Cranen and Lou Boves",

year = "2012",

month = mar,

day = "25",

doi = "10.1109/ICASSP.2012.6288831",

language = "English",

isbn = "9781467300469",

series = "ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings",

publisher = "IEEE",

pages = "4145--4148",

booktitle = "2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings",

address = "United States",

note = "2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012, ICASSP 2012 ; Conference date: 25-03-2012 Through 30-03-2012",

url = "https://doi.org/10.1109/ICASSP15465.2012",

}

Huang, H, Liu, Y, Ten Bosch, L, Cranen, B & Boves, L 2012, Knowledge-based quadratic discriminant analysis for phonetic classification. in 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings. ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings, IEEE, pp. 4145-4148, 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012, Kyoto, Japan, 25/03/12. https://doi.org/10.1109/ICASSP.2012.6288831

Knowledge-based quadratic discriminant analysis for phonetic classification. / Huang, Heyun; Liu, Yang; Ten Bosch, Louis et al.
2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings. IEEE, 2012. p. 4145-4148 (ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings).

Research output: Chapter in book/report/conference proceeding › Conference proceeding › peer-review

TY - GEN

T1 - Knowledge-based quadratic discriminant analysis for phonetic classification

AU - Huang, Heyun

AU - Liu, Yang

AU - Ten Bosch, Louis

AU - Cranen, Bert

AU - Boves, Lou

PY - 2012/3/25

Y1 - 2012/3/25

N2 - Modeling the second-order statistics of articulatory trajectories is likely to improve the performance in classifying phone segments compared to using only linear combinations of MFCCs. Nevertheless, the extremely high dimensionality of the feature space spanned by a combination of monomials of degree-1 and degree-2 makes it difficult to effectively exploit the discriminative information in the full covariance matrix. This paper proposes a novel algorithm, dubbed Knowledge-based Quadratic Discriminant Analysis (KnQDA), for reducing the number of dimensions of the space spanned by degree-1 and degree-2 monomials by using phonetic knowledge for selecting the set of degree-2 monomials that are most likely to improve classification. KnQDA seeks a trade-off between overfitting and undertraining, which further improves the learnability. Binary classifications on all pairs of phones in TIMIT show the effectiveness of the proposed method, especially on those phone pairs that overlap strongly in the linear feature space.

AB - Modeling the second-order statistics of articulatory trajectories is likely to improve the performance in classifying phone segments compared to using only linear combinations of MFCCs. Nevertheless, the extremely high dimensionality of the feature space spanned by a combination of monomials of degree-1 and degree-2 makes it difficult to effectively exploit the discriminative information in the full covariance matrix. This paper proposes a novel algorithm, dubbed Knowledge-based Quadratic Discriminant Analysis (KnQDA), for reducing the number of dimensions of the space spanned by degree-1 and degree-2 monomials by using phonetic knowledge for selecting the set of degree-2 monomials that are most likely to improve classification. KnQDA seeks a trade-off between overfitting and undertraining, which further improves the learnability. Binary classifications on all pairs of phones in TIMIT show the effectiveness of the proposed method, especially on those phone pairs that overlap strongly in the linear feature space.

KW - Dimensionality Reduction

KW - Knowledge-Based Quadratic Discriminant Analysis

KW - Phone Classification

KW - TIMIT

UR - https://www.scopus.com/pages/publications/84867593509

U2 - 10.1109/ICASSP.2012.6288831

DO - 10.1109/ICASSP.2012.6288831

M3 - Conference proceeding

AN - SCOPUS:84867593509

SN - 9781467300469

T3 - ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings

SP - 4145

EP - 4148

BT - 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings

PB - IEEE

T2 - 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012

Y2 - 25 March 2012 through 30 March 2012

ER -

Huang H, Liu Y, Ten Bosch L, Cranen B, Boves L. Knowledge-based quadratic discriminant analysis for phonetic classification. In 2012 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2012 - Proceedings. IEEE. 2012. p. 4145-4148. (ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings). doi: 10.1109/ICASSP.2012.6288831

Knowledge-based quadratic discriminant analysis for phonetic classification

Abstract

Publication series

Conference

UN SDGs

User-Defined Keywords

Access to Document

Other files and links

Fingerprint

Cite this