Using text and acoustic features in predicting glottal excitation waveforms for parametric speech synthesis with recurrent neural networks

Lauri Juvela; Xin Wang; Shinji Takaki; Manu Airaksinen; Junichi Yamagishi; Paavo Alku

doi:10.21437/Interspeech.2016-712

Using text and acoustic features in predicting glottal excitation waveforms for parametric speech synthesis with recurrent neural networks

Lauri Juvela, Xin Wang, Shinji Takaki, Manu Airaksinen, Junichi Yamagishi, Paavo Alku

Speech Communication Technology

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

6 Citations (Scopus)

Abstract

This work studies the use of deep learning methods to directly model glottal excitation waveforms from context dependent text features in a text-to-speech synthesis system. Glottal vocoding is integrated into a deep neural network-based text-to-speech framework where text and acoustic features can be flexibly used as both network inputs or outputs. Long short-term memory recurrent neural networks are utilised in two stages: first, in mapping text features to acoustic features and second, in predicting glottal waveforms from the text and/or acoustic features. Results show that using the text features directly yields similar quality to the prediction of the excitation from acoustic features, both outperforming a baseline system based on using a fixed glottal pulse for excitation generation.

Original language	English
Title of host publication	Proceedings of the Annual Conference of the International Speech Communication Association
Subtitle of host publication	Interspeech'16, San Francisco, USA, Sept. 8-12, 2016
Publisher	International Speech Communication Association (ISCA)
Pages	2283-2287
Number of pages	5
Volume	08-12-September-2016
ISBN (Electronic)	978-1-5108-3313-5
DOIs	https://doi.org/10.21437/Interspeech.2016-712
Publication status	Published - 2016
MoE publication type	A4 Conference publication
Event	Interspeech - San Francisco, United States Duration: 8 Sept 2016 → 12 Sept 2016 Conference number: 17

Publication series

Name	Proceedings of the Annual Conference of the International Speech Communication Association
Publisher	International Speech Communications Association
ISSN (Print)	1990-9770
ISSN (Electronic)	2308-457X

Conference

Conference	Interspeech
Country/Territory	United States
City	San Francisco
Period	08/09/2016 → 12/09/2016

Keywords

Excitation modelling
Glottal vocoding
LSTM
Parametric speech synthesis

Access to Document

10.21437/Interspeech.2016-712

Cite this

Juvela, L., Wang, X., Takaki, S., Airaksinen, M., Yamagishi, J., & Alku, P. (2016). Using text and acoustic features in predicting glottal excitation waveforms for parametric speech synthesis with recurrent neural networks. In Proceedings of the Annual Conference of the International Speech Communication Association: Interspeech'16, San Francisco, USA, Sept. 8-12, 2016 (Vol. 08-12-September-2016, pp. 2283-2287). (Proceedings of the Annual Conference of the International Speech Communication Association). International Speech Communication Association (ISCA). https://doi.org/10.21437/Interspeech.2016-712

Juvela, Lauri ; Wang, Xin ; Takaki, Shinji et al. / Using text and acoustic features in predicting glottal excitation waveforms for parametric speech synthesis with recurrent neural networks. Proceedings of the Annual Conference of the International Speech Communication Association: Interspeech'16, San Francisco, USA, Sept. 8-12, 2016. Vol. 08-12-September-2016 International Speech Communication Association (ISCA), 2016. pp. 2283-2287 (Proceedings of the Annual Conference of the International Speech Communication Association).

@inproceedings{121d403ebf0c4f908067e0446af01b7a,

title = "Using text and acoustic features in predicting glottal excitation waveforms for parametric speech synthesis with recurrent neural networks",

abstract = "This work studies the use of deep learning methods to directly model glottal excitation waveforms from context dependent text features in a text-to-speech synthesis system. Glottal vocoding is integrated into a deep neural network-based text-to-speech framework where text and acoustic features can be flexibly used as both network inputs or outputs. Long short-term memory recurrent neural networks are utilised in two stages: first, in mapping text features to acoustic features and second, in predicting glottal waveforms from the text and/or acoustic features. Results show that using the text features directly yields similar quality to the prediction of the excitation from acoustic features, both outperforming a baseline system based on using a fixed glottal pulse for excitation generation.",

keywords = "Excitation modelling, Glottal vocoding, LSTM, Parametric speech synthesis",

author = "Lauri Juvela and Xin Wang and Shinji Takaki and Manu Airaksinen and Junichi Yamagishi and Paavo Alku",

year = "2016",

doi = "10.21437/Interspeech.2016-712",

language = "English",

volume = "08-12-September-2016",

series = "Proceedings of the Annual Conference of the International Speech Communication Association",

publisher = "International Speech Communication Association (ISCA)",

pages = "2283--2287",

booktitle = "Proceedings of the Annual Conference of the International Speech Communication Association",

address = "France",

note = "Interspeech ; Conference date: 08-09-2016 Through 12-09-2016",

}

Juvela, L, Wang, X, Takaki, S, Airaksinen, M, Yamagishi, J & Alku, P 2016, Using text and acoustic features in predicting glottal excitation waveforms for parametric speech synthesis with recurrent neural networks. in Proceedings of the Annual Conference of the International Speech Communication Association: Interspeech'16, San Francisco, USA, Sept. 8-12, 2016. vol. 08-12-September-2016, Proceedings of the Annual Conference of the International Speech Communication Association, International Speech Communication Association (ISCA), pp. 2283-2287, Interspeech, San Francisco, California, United States, 08/09/2016. https://doi.org/10.21437/Interspeech.2016-712

Using text and acoustic features in predicting glottal excitation waveforms for parametric speech synthesis with recurrent neural networks. / Juvela, Lauri; Wang, Xin; Takaki, Shinji et al.
Proceedings of the Annual Conference of the International Speech Communication Association: Interspeech'16, San Francisco, USA, Sept. 8-12, 2016. Vol. 08-12-September-2016 International Speech Communication Association (ISCA), 2016. p. 2283-2287 (Proceedings of the Annual Conference of the International Speech Communication Association).

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

TY - GEN

T1 - Using text and acoustic features in predicting glottal excitation waveforms for parametric speech synthesis with recurrent neural networks

AU - Juvela, Lauri

AU - Wang, Xin

AU - Takaki, Shinji

AU - Airaksinen, Manu

AU - Yamagishi, Junichi

AU - Alku, Paavo

N1 - Conference code: 17

PY - 2016

Y1 - 2016

N2 - This work studies the use of deep learning methods to directly model glottal excitation waveforms from context dependent text features in a text-to-speech synthesis system. Glottal vocoding is integrated into a deep neural network-based text-to-speech framework where text and acoustic features can be flexibly used as both network inputs or outputs. Long short-term memory recurrent neural networks are utilised in two stages: first, in mapping text features to acoustic features and second, in predicting glottal waveforms from the text and/or acoustic features. Results show that using the text features directly yields similar quality to the prediction of the excitation from acoustic features, both outperforming a baseline system based on using a fixed glottal pulse for excitation generation.

AB - This work studies the use of deep learning methods to directly model glottal excitation waveforms from context dependent text features in a text-to-speech synthesis system. Glottal vocoding is integrated into a deep neural network-based text-to-speech framework where text and acoustic features can be flexibly used as both network inputs or outputs. Long short-term memory recurrent neural networks are utilised in two stages: first, in mapping text features to acoustic features and second, in predicting glottal waveforms from the text and/or acoustic features. Results show that using the text features directly yields similar quality to the prediction of the excitation from acoustic features, both outperforming a baseline system based on using a fixed glottal pulse for excitation generation.

KW - Excitation modelling

KW - Glottal vocoding

KW - LSTM

KW - Parametric speech synthesis

UR - http://www.scopus.com/inward/record.url?scp=84994228803&partnerID=8YFLogxK

U2 - 10.21437/Interspeech.2016-712

DO - 10.21437/Interspeech.2016-712

M3 - Conference article in proceedings

AN - SCOPUS:84994228803

VL - 08-12-September-2016

T3 - Proceedings of the Annual Conference of the International Speech Communication Association

SP - 2283

EP - 2287

BT - Proceedings of the Annual Conference of the International Speech Communication Association

PB - International Speech Communication Association (ISCA)

T2 - Interspeech

Y2 - 8 September 2016 through 12 September 2016

ER -

Juvela L, Wang X, Takaki S, Airaksinen M, Yamagishi J, Alku P. Using text and acoustic features in predicting glottal excitation waveforms for parametric speech synthesis with recurrent neural networks. In Proceedings of the Annual Conference of the International Speech Communication Association: Interspeech'16, San Francisco, USA, Sept. 8-12, 2016. Vol. 08-12-September-2016. International Speech Communication Association (ISCA). 2016. p. 2283-2287. (Proceedings of the Annual Conference of the International Speech Communication Association). doi: 10.21437/Interspeech.2016-712

Using text and acoustic features in predicting glottal excitation waveforms for parametric speech synthesis with recurrent neural networks

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this