Vowel duration measurement using deep neural networks

Yossi Adi; Joseph Keshet; Matthew Goldrick

doi:10.1109/MLSP.2015.7324331

Vowel duration measurement using deep neural networks

Yossi Adi
, Joseph Keshet
, Matthew Goldrick

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

10 Scopus citations

Abstract

Vowel durations are most often utilized in studies addressing specific issues in phonetics. Thus far this has been hampered by a reliance on subjective, labor-intensive manual annotation. Our goal is to build an algorithm for automatic accurate measurement of vowel duration, where the input to the algorithm is a speech segment contains one vowel preceded and followed by consonants (CVC). Our algorithm is based on a deep neural network trained at the frame level on manually annotated data from a phonetic study. Specifically, we try two deep-network architectures: convolutional neural network (CNN), and deep belief network (DBN), and compare their accuracy to an HMM-based forced aligner. Results suggest that CNN is better than DBN, and both CNN and HMM-based forced aligner are comparable in their results, but neither of them yielded the same predictions as models fit to manually annotated data.

Original language	English
Title of host publication	2015 IEEE International Workshop on Machine Learning for Signal Processing - Proceedings of MLSP 2015
Editors	Deniz Erdogmus, Serdar Kozat, Jan Larsen, Murat Akcakaya
Publisher	IEEE Computer Society
ISBN (Electronic)	9781467374545
DOIs	https://doi.org/10.1109/MLSP.2015.7324331
State	Published - 10 Nov 2015
Externally published	Yes
Event	25th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2015 - Boston, United States Duration: 17 Sep 2015 → 20 Sep 2015

Publication series

Name	IEEE International Workshop on Machine Learning for Signal Processing, MLSP
Volume	2015-November
ISSN (Print)	2161-0363
ISSN (Electronic)	2161-0371

Conference

Conference	25th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2015
Country/Territory	United States
City	Boston
Period	17/09/15 → 20/09/15

Bibliographical note

Publisher Copyright:
© 2015 IEEE.

Keywords

Forced alignment
convolution neural networks
deep belief networks
hidden Markov models
vowel duration measurement

Access to Document

10.1109/MLSP.2015.7324331

Cite this

Adi, Y., Keshet, J., & Goldrick, M. (2015). Vowel duration measurement using deep neural networks. In D. Erdogmus, S. Kozat, J. Larsen, & M. Akcakaya (Eds.), 2015 IEEE International Workshop on Machine Learning for Signal Processing - Proceedings of MLSP 2015 Article 7324331 (IEEE International Workshop on Machine Learning for Signal Processing, MLSP; Vol. 2015-November). IEEE Computer Society. https://doi.org/10.1109/MLSP.2015.7324331

Adi, Yossi ; Keshet, Joseph ; Goldrick, Matthew. / Vowel duration measurement using deep neural networks. 2015 IEEE International Workshop on Machine Learning for Signal Processing - Proceedings of MLSP 2015. editor / Deniz Erdogmus ; Serdar Kozat ; Jan Larsen ; Murat Akcakaya. IEEE Computer Society, 2015. (IEEE International Workshop on Machine Learning for Signal Processing, MLSP).

@inproceedings{e2b5781f081c48e88dbab1ed7443dd23,

title = "Vowel duration measurement using deep neural networks",

abstract = "Vowel durations are most often utilized in studies addressing specific issues in phonetics. Thus far this has been hampered by a reliance on subjective, labor-intensive manual annotation. Our goal is to build an algorithm for automatic accurate measurement of vowel duration, where the input to the algorithm is a speech segment contains one vowel preceded and followed by consonants (CVC). Our algorithm is based on a deep neural network trained at the frame level on manually annotated data from a phonetic study. Specifically, we try two deep-network architectures: convolutional neural network (CNN), and deep belief network (DBN), and compare their accuracy to an HMM-based forced aligner. Results suggest that CNN is better than DBN, and both CNN and HMM-based forced aligner are comparable in their results, but neither of them yielded the same predictions as models fit to manually annotated data.",

keywords = "Forced alignment, convolution neural networks, deep belief networks, hidden Markov models, vowel duration measurement",

author = "Yossi Adi and Joseph Keshet and Matthew Goldrick",

note = "Publisher Copyright: {\textcopyright} 2015 IEEE.; 25th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2015 ; Conference date: 17-09-2015 Through 20-09-2015",

year = "2015",

month = nov,

day = "10",

doi = "10.1109/MLSP.2015.7324331",

language = "אנגלית",

series = "IEEE International Workshop on Machine Learning for Signal Processing, MLSP",

publisher = "IEEE Computer Society",

editor = "Deniz Erdogmus and Serdar Kozat and Jan Larsen and Murat Akcakaya",

booktitle = "2015 IEEE International Workshop on Machine Learning for Signal Processing - Proceedings of MLSP 2015",

address = "ארצות הברית",

}

Adi, Y, Keshet, J & Goldrick, M 2015, Vowel duration measurement using deep neural networks. in D Erdogmus, S Kozat, J Larsen & M Akcakaya (eds), 2015 IEEE International Workshop on Machine Learning for Signal Processing - Proceedings of MLSP 2015., 7324331, IEEE International Workshop on Machine Learning for Signal Processing, MLSP, vol. 2015-November, IEEE Computer Society, 25th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2015, Boston, United States, 17/09/15. https://doi.org/10.1109/MLSP.2015.7324331

Vowel duration measurement using deep neural networks. / Adi, Yossi; Keshet, Joseph; Goldrick, Matthew.
2015 IEEE International Workshop on Machine Learning for Signal Processing - Proceedings of MLSP 2015. ed. / Deniz Erdogmus; Serdar Kozat; Jan Larsen; Murat Akcakaya. IEEE Computer Society, 2015. 7324331 (IEEE International Workshop on Machine Learning for Signal Processing, MLSP; Vol. 2015-November).

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

TY - GEN

T1 - Vowel duration measurement using deep neural networks

AU - Adi, Yossi

AU - Keshet, Joseph

AU - Goldrick, Matthew

PY - 2015/11/10

Y1 - 2015/11/10

N2 - Vowel durations are most often utilized in studies addressing specific issues in phonetics. Thus far this has been hampered by a reliance on subjective, labor-intensive manual annotation. Our goal is to build an algorithm for automatic accurate measurement of vowel duration, where the input to the algorithm is a speech segment contains one vowel preceded and followed by consonants (CVC). Our algorithm is based on a deep neural network trained at the frame level on manually annotated data from a phonetic study. Specifically, we try two deep-network architectures: convolutional neural network (CNN), and deep belief network (DBN), and compare their accuracy to an HMM-based forced aligner. Results suggest that CNN is better than DBN, and both CNN and HMM-based forced aligner are comparable in their results, but neither of them yielded the same predictions as models fit to manually annotated data.

AB - Vowel durations are most often utilized in studies addressing specific issues in phonetics. Thus far this has been hampered by a reliance on subjective, labor-intensive manual annotation. Our goal is to build an algorithm for automatic accurate measurement of vowel duration, where the input to the algorithm is a speech segment contains one vowel preceded and followed by consonants (CVC). Our algorithm is based on a deep neural network trained at the frame level on manually annotated data from a phonetic study. Specifically, we try two deep-network architectures: convolutional neural network (CNN), and deep belief network (DBN), and compare their accuracy to an HMM-based forced aligner. Results suggest that CNN is better than DBN, and both CNN and HMM-based forced aligner are comparable in their results, but neither of them yielded the same predictions as models fit to manually annotated data.

KW - Forced alignment

KW - convolution neural networks

KW - deep belief networks

KW - hidden Markov models

KW - vowel duration measurement

UR - https://www.scopus.com/pages/publications/84960910537

U2 - 10.1109/MLSP.2015.7324331

DO - 10.1109/MLSP.2015.7324331

M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???

AN - SCOPUS:84960910537

T3 - IEEE International Workshop on Machine Learning for Signal Processing, MLSP

BT - 2015 IEEE International Workshop on Machine Learning for Signal Processing - Proceedings of MLSP 2015

A2 - Erdogmus, Deniz

A2 - Kozat, Serdar

A2 - Larsen, Jan

A2 - Akcakaya, Murat

PB - IEEE Computer Society

T2 - 25th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2015

Y2 - 17 September 2015 through 20 September 2015

ER -

Adi Y, Keshet J, Goldrick M. Vowel duration measurement using deep neural networks. In Erdogmus D, Kozat S, Larsen J, Akcakaya M, editors, 2015 IEEE International Workshop on Machine Learning for Signal Processing - Proceedings of MLSP 2015. IEEE Computer Society. 2015. 7324331. (IEEE International Workshop on Machine Learning for Signal Processing, MLSP). doi: 10.1109/MLSP.2015.7324331

Vowel duration measurement using deep neural networks

Abstract

Publication series

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this