An unsupervised method to select a speaker subset from large multi-speaker speech synthesis datasets

Pilar Oplustil Gallegos; Jennifer Williams; Joanna Rownicka; Simon King

doi:10.21437/Interspeech.2020-2567

An unsupervised method to select a speaker subset from large multi-speaker speech synthesis datasets

Pilar Oplustil Gallegos^*, Jennifer Williams, Joanna Rownicka, Simon King

^*Corresponding author for this work

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

Abstract / Description of output

Large multi-speaker datasets for TTS typically contain diverse speakers, recording conditions, styles and quality of data. Although one might generally presume that more data is better, in this paper we show that a model trained on a carefully-chosen subset of speakers from LibriTTS provides significantly better quality synthetic speech than a model trained on a larger set. We propose an unsupervised methodology to find this subset by clustering per-speaker acoustic representations.

Original language	English
Title of host publication	Proceedings of the Annual Conference of the International Speech Communication Association
Pages	1758-1762
Number of pages	5
Volume	2020-October
DOIs	https://doi.org/10.21437/Interspeech.2020-2567
Publication status	Published - 29 Oct 2020
Event	21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020 - Shanghai, China Duration: 25 Oct 2020 → 29 Oct 2020

Publication series

Name	Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
ISSN (Print)	2308-457X

Conference

Conference	21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020
Country/Territory	China
City	Shanghai
Period	25/10/20 → 29/10/20

Keywords / Materials (for Non-textual outputs)

clustering
data
multi-speaker
sequence-to-sequence models
speaker representation
speech synthesis

Access to Document

10.21437/Interspeech.2020-2567

GallegosEtal2020AnUnsupervisedMethodAccepted author manuscript, 581 KB

https://www.isca-speech.org/archive/Interspeech_2020/pdfs/2567.pdf

Cite this

Gallegos, P. O., Williams, J., Rownicka, J., & King, S. (2020). An unsupervised method to select a speaker subset from large multi-speaker speech synthesis datasets. In Proceedings of the Annual Conference of the International Speech Communication Association (Vol. 2020-October, pp. 1758-1762). (Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH). https://doi.org/10.21437/Interspeech.2020-2567

Gallegos, Pilar Oplustil ; Williams, Jennifer ; Rownicka, Joanna et al. / An unsupervised method to select a speaker subset from large multi-speaker speech synthesis datasets. Proceedings of the Annual Conference of the International Speech Communication Association. Vol. 2020-October 2020. pp. 1758-1762 (Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH).

@inproceedings{f22287ce65084603869c77664ac658da,

title = "An unsupervised method to select a speaker subset from large multi-speaker speech synthesis datasets",

abstract = "Large multi-speaker datasets for TTS typically contain diverse speakers, recording conditions, styles and quality of data. Although one might generally presume that more data is better, in this paper we show that a model trained on a carefully-chosen subset of speakers from LibriTTS provides significantly better quality synthetic speech than a model trained on a larger set. We propose an unsupervised methodology to find this subset by clustering per-speaker acoustic representations.",

keywords = "clustering, data, multi-speaker, sequence-to-sequence models, speaker representation, speech synthesis",

author = "Gallegos, {Pilar Oplustil} and Jennifer Williams and Joanna Rownicka and Simon King",

note = "Funding Information: this work was supported in part by: ANID, Becas Chile, n?72190135; the EPSRC Centre for Doctoral Training in Data Science, funded by the UK Engineering and Physical Sciences Research Council (grant EP/L016427/1) and the University of Edinburgh; and a PhD studentship from the DataLab Innovation Centre, Ericsson Media Services, and Quorate Technology.; 21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020 ; Conference date: 25-10-2020 Through 29-10-2020",

year = "2020",

month = oct,

day = "29",

doi = "10.21437/Interspeech.2020-2567",

language = "English",

volume = "2020-October",

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

pages = "1758--1762",

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

}

Gallegos, PO, Williams, J, Rownicka, J & King, S 2020, An unsupervised method to select a speaker subset from large multi-speaker speech synthesis datasets. in Proceedings of the Annual Conference of the International Speech Communication Association. vol. 2020-October, Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH, pp. 1758-1762, 21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020, Shanghai, China, 25/10/20. https://doi.org/10.21437/Interspeech.2020-2567

An unsupervised method to select a speaker subset from large multi-speaker speech synthesis datasets. / Gallegos, Pilar Oplustil; Williams, Jennifer; Rownicka, Joanna et al.
Proceedings of the Annual Conference of the International Speech Communication Association. Vol. 2020-October 2020. p. 1758-1762 (Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH).

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

TY - GEN

T1 - An unsupervised method to select a speaker subset from large multi-speaker speech synthesis datasets

AU - Gallegos, Pilar Oplustil

AU - Williams, Jennifer

AU - Rownicka, Joanna

AU - King, Simon

N1 - Funding Information: this work was supported in part by: ANID, Becas Chile, n?72190135; the EPSRC Centre for Doctoral Training in Data Science, funded by the UK Engineering and Physical Sciences Research Council (grant EP/L016427/1) and the University of Edinburgh; and a PhD studentship from the DataLab Innovation Centre, Ericsson Media Services, and Quorate Technology.

PY - 2020/10/29

Y1 - 2020/10/29

N2 - Large multi-speaker datasets for TTS typically contain diverse speakers, recording conditions, styles and quality of data. Although one might generally presume that more data is better, in this paper we show that a model trained on a carefully-chosen subset of speakers from LibriTTS provides significantly better quality synthetic speech than a model trained on a larger set. We propose an unsupervised methodology to find this subset by clustering per-speaker acoustic representations.

AB - Large multi-speaker datasets for TTS typically contain diverse speakers, recording conditions, styles and quality of data. Although one might generally presume that more data is better, in this paper we show that a model trained on a carefully-chosen subset of speakers from LibriTTS provides significantly better quality synthetic speech than a model trained on a larger set. We propose an unsupervised methodology to find this subset by clustering per-speaker acoustic representations.

KW - clustering

KW - data

KW - multi-speaker

KW - sequence-to-sequence models

KW - speaker representation

KW - speech synthesis

U2 - 10.21437/Interspeech.2020-2567

DO - 10.21437/Interspeech.2020-2567

M3 - Conference contribution

AN - SCOPUS:85098189222

VL - 2020-October

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

SP - 1758

EP - 1762

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

T2 - 21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020

Y2 - 25 October 2020 through 29 October 2020

ER -

Gallegos PO, Williams J, Rownicka J, King S. An unsupervised method to select a speaker subset from large multi-speaker speech synthesis datasets. In Proceedings of the Annual Conference of the International Speech Communication Association. Vol. 2020-October. 2020. p. 1758-1762. (Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH). doi: 10.21437/Interspeech.2020-2567

An unsupervised method to select a speaker subset from large multi-speaker speech synthesis datasets

Abstract / Description of output

Publication series

Conference

Keywords / Materials (for Non-textual outputs)

Access to Document

Fingerprint

Cite this