Blind phoneme segmentation with temporal prediction errors

Paul Michel; Okko Räsänen; Roland Thiolliere; Emmanuel Dupoux

doi:10.18653/v1/P17-3011

Blind phoneme segmentation with temporal prediction errors

Paul Michel, Okko Räsänen, Roland Thiolliere, Emmanuel Dupoux

Department of Signal Processing and Acoustics

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

4 Citations (Scopus)

118 Downloads (Pure)

Abstract

Phonemic segmentation of speech is a critical step of speech recognition systems. We propose a novel unsupervised algorithm based on sequence prediction models such as Markov chains and recurrent neural networks. Our approach consists in analyzing the error profile of a model trained to predict speech features frame-by-frame. Specifically, we try to learn the dynamics of speech in the MFCC space and hypothesize boundaries from local maxima in the prediction error. We evaluate our system on the TIMIT dataset, with improvements over similar methods.

Original language	English
Title of host publication	Proceedings of the Student Research Workshop at the Annual Meeting of the Association for Computational Linguistics
Pages	62-68
Number of pages	7
ISBN (Electronic)	978-1-945626-56-2
DOIs	https://doi.org/10.18653/v1/P17-3011
Publication status	Published - 2017
MoE publication type	A4 Article in a conference publication
Event	Annual Meeting of the Association for Computational Linguistics: Student Research Workshop - Vancouver, Canada Duration: 30 Jul 2017 → 4 Aug 2017

Workshop

Workshop	Annual Meeting of the Association for Computational Linguistics
Abbreviated title	SRW
Country/Territory	Canada
City	Vancouver
Period	30/07/2017 → 04/08/2017

Access to Document

10.18653/v1/P17-3011Licence: CC BY

ELEC-Michel-et-al-Blind phoneme_ACL2017Final published version, 696 KBLicence: CC BY

Cite this

@inproceedings{fe9e8a22223f4ad29bb6685df4282cbb,

title = "Blind phoneme segmentation with temporal prediction errors",

abstract = "Phonemic segmentation of speech is a critical step of speech recognition systems. We propose a novel unsupervised algorithm based on sequence prediction models such as Markov chains and recurrent neural networks. Our approach consists in analyzing the error profile of a model trained to predict speech features frame-by-frame. Specifically, we try to learn the dynamics of speech in the MFCC space and hypothesize boundaries from local maxima in the prediction error. We evaluate our system on the TIMIT dataset, with improvements over similar methods.",

author = "Paul Michel and Okko R{\"a}s{\"a}nen and Roland Thiolliere and Emmanuel Dupoux",

year = "2017",

doi = "10.18653/v1/P17-3011",

language = "English",

pages = "62--68",

booktitle = "Proceedings of the Student Research Workshop at the Annual Meeting of the Association for Computational Linguistics",

note = "Annual Meeting of the Association for Computational Linguistics : Student Research Workshop, SRW ; Conference date: 30-07-2017 Through 04-08-2017",

}

Michel, P, Räsänen, O, Thiolliere, R & Dupoux, E 2017, Blind phoneme segmentation with temporal prediction errors. in Proceedings of the Student Research Workshop at the Annual Meeting of the Association for Computational Linguistics. pp. 62-68, Annual Meeting of the Association for Computational Linguistics, Vancouver, Canada, 30/07/2017. https://doi.org/10.18653/v1/P17-3011

Blind phoneme segmentation with temporal prediction errors. / Michel, Paul; Räsänen, Okko; Thiolliere, Roland; Dupoux, Emmanuel.

Proceedings of the Student Research Workshop at the Annual Meeting of the Association for Computational Linguistics. 2017. p. 62-68.

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

TY - GEN

T1 - Blind phoneme segmentation with temporal prediction errors

AU - Michel, Paul

AU - Räsänen, Okko

AU - Thiolliere, Roland

AU - Dupoux, Emmanuel

PY - 2017

Y1 - 2017

N2 - Phonemic segmentation of speech is a critical step of speech recognition systems. We propose a novel unsupervised algorithm based on sequence prediction models such as Markov chains and recurrent neural networks. Our approach consists in analyzing the error profile of a model trained to predict speech features frame-by-frame. Specifically, we try to learn the dynamics of speech in the MFCC space and hypothesize boundaries from local maxima in the prediction error. We evaluate our system on the TIMIT dataset, with improvements over similar methods.

AB - Phonemic segmentation of speech is a critical step of speech recognition systems. We propose a novel unsupervised algorithm based on sequence prediction models such as Markov chains and recurrent neural networks. Our approach consists in analyzing the error profile of a model trained to predict speech features frame-by-frame. Specifically, we try to learn the dynamics of speech in the MFCC space and hypothesize boundaries from local maxima in the prediction error. We evaluate our system on the TIMIT dataset, with improvements over similar methods.

U2 - 10.18653/v1/P17-3011

DO - 10.18653/v1/P17-3011

M3 - Conference contribution

SP - 62

EP - 68

BT - Proceedings of the Student Research Workshop at the Annual Meeting of the Association for Computational Linguistics

T2 - Annual Meeting of the Association for Computational Linguistics

Y2 - 30 July 2017 through 4 August 2017

ER -

Blind phoneme segmentation with temporal prediction errors

Abstract

Workshop

Access to Document

Fingerprint

Cite this