Investigating Manifold Learning Technique for Robust Speech Recognition

Bi Cheng Yan, Chin Hong Shih, Berlin Chen, Shih Hung Liu

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Developing robustness methods is imperative to retaining good performance for automatic speech recognition (ASR)systems when being confronted with different environmental noise or channel distortion. Previous studies have pointed out that exploration of low-dimensional structures of speech features is beneficial to generating robust features so as to enhance ASR performance. Along this research direction, we argue that the intrinsic structures of speech features lying on a manifold subspace of low dimensionality residing in their original ambient space of high dimensionality. This way, noise components can be ruled out by projecting noisy speech features into the pre-learned subspace of manifold structures. This paper explores the intrinsic geometric low-dimensional manifold structures inherent speech features' modulation spectra, with the goal to generate speech features that are more robust to environmental noise and channel distortion. The key novelty of our work is two-fold: 1)we put forward an innovative use of the graph-regularization based method to generate robust speech features by preserving the inherent manifold structures of modulation spectra and excluding irrelevant ones, and 2)we also compare our approach with several mainstream methods that also explores low-dimensional structures of data instances with in-depth analysis. A comprehensive set of empirical experiments carried out on an ASR benchmark task seem to reveal the superior performance of our proposed methods.

Original languageEnglish
Title of host publicationProceedings of the 2018 International Conference on Asian Language Processing, IALP 2018
EditorsMinghui Dong, Fariska Z. Ruskanda, Herry Sujaini, Ade Romadhony, Moch. Bijaksana, Elvira Nurfadhilah, Lyla Ruslana Aini, Arif Bijaksana Putra Negara
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages68-73
Number of pages6
ISBN (Electronic)9781728111766
DOIs
Publication statusPublished - 2019 Jan 28
Event22nd International Conference on Asian Language Processing, IALP 2018 - Bandung, Indonesia
Duration: 2018 Nov 152018 Nov 17

Publication series

NameProceedings of the 2018 International Conference on Asian Language Processing, IALP 2018

Conference

Conference22nd International Conference on Asian Language Processing, IALP 2018
CountryIndonesia
CityBandung
Period18/11/1518/11/17

Fingerprint

Speech recognition
learning
Modulation
performance
Speech Recognition
Automatic Speech Recognition
Experiments
experiment

Keywords

  • automatic speech recognition
  • low-dimensional structures
  • manifold learning
  • robustness

ASJC Scopus subject areas

  • Language and Linguistics
  • Linguistics and Language
  • Computer Science Applications

Cite this

Yan, B. C., Shih, C. H., Chen, B., & Liu, S. H. (2019). Investigating Manifold Learning Technique for Robust Speech Recognition. In M. Dong, F. Z. Ruskanda, H. Sujaini, A. Romadhony, M. Bijaksana, E. Nurfadhilah, L. R. Aini, ... A. B. P. Negara (Eds.), Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018 (pp. 68-73). [8629269] (Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/IALP.2018.8629269

Investigating Manifold Learning Technique for Robust Speech Recognition. / Yan, Bi Cheng; Shih, Chin Hong; Chen, Berlin; Liu, Shih Hung.

Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018. ed. / Minghui Dong; Fariska Z. Ruskanda; Herry Sujaini; Ade Romadhony; Moch. Bijaksana; Elvira Nurfadhilah; Lyla Ruslana Aini; Arif Bijaksana Putra Negara. Institute of Electrical and Electronics Engineers Inc., 2019. p. 68-73 8629269 (Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Yan, BC, Shih, CH, Chen, B & Liu, SH 2019, Investigating Manifold Learning Technique for Robust Speech Recognition. in M Dong, FZ Ruskanda, H Sujaini, A Romadhony, M Bijaksana, E Nurfadhilah, LR Aini & ABP Negara (eds), Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018., 8629269, Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018, Institute of Electrical and Electronics Engineers Inc., pp. 68-73, 22nd International Conference on Asian Language Processing, IALP 2018, Bandung, Indonesia, 18/11/15. https://doi.org/10.1109/IALP.2018.8629269
Yan BC, Shih CH, Chen B, Liu SH. Investigating Manifold Learning Technique for Robust Speech Recognition. In Dong M, Ruskanda FZ, Sujaini H, Romadhony A, Bijaksana M, Nurfadhilah E, Aini LR, Negara ABP, editors, Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018. Institute of Electrical and Electronics Engineers Inc. 2019. p. 68-73. 8629269. (Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018). https://doi.org/10.1109/IALP.2018.8629269
Yan, Bi Cheng ; Shih, Chin Hong ; Chen, Berlin ; Liu, Shih Hung. / Investigating Manifold Learning Technique for Robust Speech Recognition. Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018. editor / Minghui Dong ; Fariska Z. Ruskanda ; Herry Sujaini ; Ade Romadhony ; Moch. Bijaksana ; Elvira Nurfadhilah ; Lyla Ruslana Aini ; Arif Bijaksana Putra Negara. Institute of Electrical and Electronics Engineers Inc., 2019. pp. 68-73 (Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018).
@inproceedings{5635b3c2aba34c7cb1c2a1459df97547,
title = "Investigating Manifold Learning Technique for Robust Speech Recognition",
abstract = "Developing robustness methods is imperative to retaining good performance for automatic speech recognition (ASR)systems when being confronted with different environmental noise or channel distortion. Previous studies have pointed out that exploration of low-dimensional structures of speech features is beneficial to generating robust features so as to enhance ASR performance. Along this research direction, we argue that the intrinsic structures of speech features lying on a manifold subspace of low dimensionality residing in their original ambient space of high dimensionality. This way, noise components can be ruled out by projecting noisy speech features into the pre-learned subspace of manifold structures. This paper explores the intrinsic geometric low-dimensional manifold structures inherent speech features' modulation spectra, with the goal to generate speech features that are more robust to environmental noise and channel distortion. The key novelty of our work is two-fold: 1)we put forward an innovative use of the graph-regularization based method to generate robust speech features by preserving the inherent manifold structures of modulation spectra and excluding irrelevant ones, and 2)we also compare our approach with several mainstream methods that also explores low-dimensional structures of data instances with in-depth analysis. A comprehensive set of empirical experiments carried out on an ASR benchmark task seem to reveal the superior performance of our proposed methods.",
keywords = "automatic speech recognition, low-dimensional structures, manifold learning, robustness",
author = "Yan, {Bi Cheng} and Shih, {Chin Hong} and Berlin Chen and Liu, {Shih Hung}",
year = "2019",
month = "1",
day = "28",
doi = "10.1109/IALP.2018.8629269",
language = "English",
series = "Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "68--73",
editor = "Minghui Dong and Ruskanda, {Fariska Z.} and Herry Sujaini and Ade Romadhony and Moch. Bijaksana and Elvira Nurfadhilah and Aini, {Lyla Ruslana} and Negara, {Arif Bijaksana Putra}",
booktitle = "Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018",

}

TY - GEN

T1 - Investigating Manifold Learning Technique for Robust Speech Recognition

AU - Yan, Bi Cheng

AU - Shih, Chin Hong

AU - Chen, Berlin

AU - Liu, Shih Hung

PY - 2019/1/28

Y1 - 2019/1/28

N2 - Developing robustness methods is imperative to retaining good performance for automatic speech recognition (ASR)systems when being confronted with different environmental noise or channel distortion. Previous studies have pointed out that exploration of low-dimensional structures of speech features is beneficial to generating robust features so as to enhance ASR performance. Along this research direction, we argue that the intrinsic structures of speech features lying on a manifold subspace of low dimensionality residing in their original ambient space of high dimensionality. This way, noise components can be ruled out by projecting noisy speech features into the pre-learned subspace of manifold structures. This paper explores the intrinsic geometric low-dimensional manifold structures inherent speech features' modulation spectra, with the goal to generate speech features that are more robust to environmental noise and channel distortion. The key novelty of our work is two-fold: 1)we put forward an innovative use of the graph-regularization based method to generate robust speech features by preserving the inherent manifold structures of modulation spectra and excluding irrelevant ones, and 2)we also compare our approach with several mainstream methods that also explores low-dimensional structures of data instances with in-depth analysis. A comprehensive set of empirical experiments carried out on an ASR benchmark task seem to reveal the superior performance of our proposed methods.

AB - Developing robustness methods is imperative to retaining good performance for automatic speech recognition (ASR)systems when being confronted with different environmental noise or channel distortion. Previous studies have pointed out that exploration of low-dimensional structures of speech features is beneficial to generating robust features so as to enhance ASR performance. Along this research direction, we argue that the intrinsic structures of speech features lying on a manifold subspace of low dimensionality residing in their original ambient space of high dimensionality. This way, noise components can be ruled out by projecting noisy speech features into the pre-learned subspace of manifold structures. This paper explores the intrinsic geometric low-dimensional manifold structures inherent speech features' modulation spectra, with the goal to generate speech features that are more robust to environmental noise and channel distortion. The key novelty of our work is two-fold: 1)we put forward an innovative use of the graph-regularization based method to generate robust speech features by preserving the inherent manifold structures of modulation spectra and excluding irrelevant ones, and 2)we also compare our approach with several mainstream methods that also explores low-dimensional structures of data instances with in-depth analysis. A comprehensive set of empirical experiments carried out on an ASR benchmark task seem to reveal the superior performance of our proposed methods.

KW - automatic speech recognition

KW - low-dimensional structures

KW - manifold learning

KW - robustness

UR - http://www.scopus.com/inward/record.url?scp=85062784456&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85062784456&partnerID=8YFLogxK

U2 - 10.1109/IALP.2018.8629269

DO - 10.1109/IALP.2018.8629269

M3 - Conference contribution

T3 - Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018

SP - 68

EP - 73

BT - Proceedings of the 2018 International Conference on Asian Language Processing, IALP 2018

A2 - Dong, Minghui

A2 - Ruskanda, Fariska Z.

A2 - Sujaini, Herry

A2 - Romadhony, Ade

A2 - Bijaksana, Moch.

A2 - Nurfadhilah, Elvira

A2 - Aini, Lyla Ruslana

A2 - Negara, Arif Bijaksana Putra

PB - Institute of Electrical and Electronics Engineers Inc.

ER -