Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech

Although recent neural text-to-speech (TTS) systems have achieved high-quality speech synthesis, there are cases where a TTS system generates low-quality speech, mainly caused by limited training data or information loss during knowledge distillation. Therefore, we propose a novel method to improve...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE access 2022, Vol.10, p.52621-52629
Hauptverfasser: Choi, Yeunju, Jung, Youngmoon, Suh, Youngjoo, Kim, Hoirin
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page 52629
container_issue
container_start_page 52621
container_title IEEE access
container_volume 10
creator Choi, Yeunju
Jung, Youngmoon
Suh, Youngjoo
Kim, Hoirin
description Although recent neural text-to-speech (TTS) systems have achieved high-quality speech synthesis, there are cases where a TTS system generates low-quality speech, mainly caused by limited training data or information loss during knowledge distillation. Therefore, we propose a novel method to improve speech quality by training a TTS model under the supervision of perceptual loss, which measures the distance between the maximum possible speech quality score and the predicted one. We first pre-train a mean opinion score (MOS) prediction model and then train a TTS model to maximize the MOS of synthesized speech using the pre-trained MOS prediction model. The proposed method can be applied independently regardless of the TTS model architecture or the cause of speech quality degradation and efficiently without increasing the inference time or model complexity. The evaluation results for the MOS and phone error rate demonstrate that our proposed approach improves previous models in terms of both naturalness and intelligibility.
doi_str_mv 10.1109/ACCESS.2022.3175810
format Article
fullrecord <record><control><sourceid>proquest_ieee_</sourceid><recordid>TN_cdi_ieee_primary_9775804</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><ieee_id>9775804</ieee_id><doaj_id>oai_doaj_org_article_94c605add4db4b9fa8fb5f63361d4929</doaj_id><sourcerecordid>2669159184</sourcerecordid><originalsourceid>FETCH-LOGICAL-c474t-e6e294493123a569b0e2b3a17bfb84577172a3ad0c6102b4cea7af63282e5c33</originalsourceid><addsrcrecordid>eNpNUU1P20AQtVArEQG_gMtKnJ3u93qPUUgLUoBWSU89rMbrMWzkZMN6IxF-fZ0aoc5lRm_eezPSK4prRqeMUfttNp8vVqspp5xPBTOqYvSsmHCmbSmU0F_-m8-Lq77f0KGqAVJmUvxZIqRd2D2THMkDvIVteEey2iP6F_LrAF3IR3IbEvrcHcnv_sR8eFqRnwmb4HOIO9LGRB7xkKAja3zLZY7lqL8svrbQ9Xj10S-K9ffFen5XLp9-3M9ny9JLI3OJGrmV0grGBShta4q8FsBM3daVVMYww0FAQ71mlNfSIxhoteAVR-WFuCjuR9smwsbtU9hCOroIwf0DYnp2kHLwHTorvaYKmkY2taxtC1Vbq8FKaNZIy-3gdTN67VN8PWCf3SYe0m743nGtLVOWVXJgiZHlU-z7hO3nVUbdKRM3ZuJOmbiPTAbV9agKiPipsGbYUin-AmOwhsY</addsrcrecordid><sourcetype>Open Website</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2669159184</pqid></control><display><type>article</type><title>Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech</title><source>IEEE Open Access Journals</source><source>DOAJ Directory of Open Access Journals</source><source>Elektronische Zeitschriftenbibliothek - Frei zugängliche E-Journals</source><creator>Choi, Yeunju ; Jung, Youngmoon ; Suh, Youngjoo ; Kim, Hoirin</creator><creatorcontrib>Choi, Yeunju ; Jung, Youngmoon ; Suh, Youngjoo ; Kim, Hoirin</creatorcontrib><description>Although recent neural text-to-speech (TTS) systems have achieved high-quality speech synthesis, there are cases where a TTS system generates low-quality speech, mainly caused by limited training data or information loss during knowledge distillation. Therefore, we propose a novel method to improve speech quality by training a TTS model under the supervision of perceptual loss, which measures the distance between the maximum possible speech quality score and the predicted one. We first pre-train a mean opinion score (MOS) prediction model and then train a TTS model to maximize the MOS of synthesized speech using the pre-trained MOS prediction model. The proposed method can be applied independently regardless of the TTS model architecture or the cause of speech quality degradation and efficiently without increasing the inference time or model complexity. The evaluation results for the MOS and phone error rate demonstrate that our proposed approach improves previous models in terms of both naturalness and intelligibility.</description><identifier>ISSN: 2169-3536</identifier><identifier>EISSN: 2169-3536</identifier><identifier>DOI: 10.1109/ACCESS.2022.3175810</identifier><identifier>CODEN: IAECCG</identifier><language>eng</language><publisher>Piscataway: IEEE</publisher><subject>Data models ; Distillation ; Intelligibility ; MOS prediction ; neural TTS ; perceptual loss ; Prediction models ; Predictions ; Predictive models ; Speech ; Speech recognition ; Speech synthesis ; Task analysis ; Training ; Training data ; Transformers</subject><ispartof>IEEE access, 2022, Vol.10, p.52621-52629</ispartof><rights>Copyright The Institute of Electrical and Electronics Engineers, Inc. (IEEE) 2022</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><citedby>FETCH-LOGICAL-c474t-e6e294493123a569b0e2b3a17bfb84577172a3ad0c6102b4cea7af63282e5c33</citedby><cites>FETCH-LOGICAL-c474t-e6e294493123a569b0e2b3a17bfb84577172a3ad0c6102b4cea7af63282e5c33</cites><orcidid>0000-0002-8787-6982 ; 0000-0003-2192-2680 ; 0000-0002-4321-379X</orcidid></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://ieeexplore.ieee.org/document/9775804$$EHTML$$P50$$Gieee$$Hfree_for_read</linktohtml><link.rule.ids>314,780,784,864,2102,4024,27633,27923,27924,27925,54933</link.rule.ids></links><search><creatorcontrib>Choi, Yeunju</creatorcontrib><creatorcontrib>Jung, Youngmoon</creatorcontrib><creatorcontrib>Suh, Youngjoo</creatorcontrib><creatorcontrib>Kim, Hoirin</creatorcontrib><title>Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech</title><title>IEEE access</title><addtitle>Access</addtitle><description>Although recent neural text-to-speech (TTS) systems have achieved high-quality speech synthesis, there are cases where a TTS system generates low-quality speech, mainly caused by limited training data or information loss during knowledge distillation. Therefore, we propose a novel method to improve speech quality by training a TTS model under the supervision of perceptual loss, which measures the distance between the maximum possible speech quality score and the predicted one. We first pre-train a mean opinion score (MOS) prediction model and then train a TTS model to maximize the MOS of synthesized speech using the pre-trained MOS prediction model. The proposed method can be applied independently regardless of the TTS model architecture or the cause of speech quality degradation and efficiently without increasing the inference time or model complexity. The evaluation results for the MOS and phone error rate demonstrate that our proposed approach improves previous models in terms of both naturalness and intelligibility.</description><subject>Data models</subject><subject>Distillation</subject><subject>Intelligibility</subject><subject>MOS prediction</subject><subject>neural TTS</subject><subject>perceptual loss</subject><subject>Prediction models</subject><subject>Predictions</subject><subject>Predictive models</subject><subject>Speech</subject><subject>Speech recognition</subject><subject>Speech synthesis</subject><subject>Task analysis</subject><subject>Training</subject><subject>Training data</subject><subject>Transformers</subject><issn>2169-3536</issn><issn>2169-3536</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2022</creationdate><recordtype>article</recordtype><sourceid>ESBDL</sourceid><sourceid>RIE</sourceid><sourceid>DOA</sourceid><recordid>eNpNUU1P20AQtVArEQG_gMtKnJ3u93qPUUgLUoBWSU89rMbrMWzkZMN6IxF-fZ0aoc5lRm_eezPSK4prRqeMUfttNp8vVqspp5xPBTOqYvSsmHCmbSmU0F_-m8-Lq77f0KGqAVJmUvxZIqRd2D2THMkDvIVteEey2iP6F_LrAF3IR3IbEvrcHcnv_sR8eFqRnwmb4HOIO9LGRB7xkKAja3zLZY7lqL8svrbQ9Xj10S-K9ffFen5XLp9-3M9ny9JLI3OJGrmV0grGBShta4q8FsBM3daVVMYww0FAQ71mlNfSIxhoteAVR-WFuCjuR9smwsbtU9hCOroIwf0DYnp2kHLwHTorvaYKmkY2taxtC1Vbq8FKaNZIy-3gdTN67VN8PWCf3SYe0m743nGtLVOWVXJgiZHlU-z7hO3nVUbdKRM3ZuJOmbiPTAbV9agKiPipsGbYUin-AmOwhsY</recordid><startdate>2022</startdate><enddate>2022</enddate><creator>Choi, Yeunju</creator><creator>Jung, Youngmoon</creator><creator>Suh, Youngjoo</creator><creator>Kim, Hoirin</creator><general>IEEE</general><general>The Institute of Electrical and Electronics Engineers, Inc. (IEEE)</general><scope>97E</scope><scope>ESBDL</scope><scope>RIA</scope><scope>RIE</scope><scope>AAYXX</scope><scope>CITATION</scope><scope>7SC</scope><scope>7SP</scope><scope>7SR</scope><scope>8BQ</scope><scope>8FD</scope><scope>JG9</scope><scope>JQ2</scope><scope>L7M</scope><scope>L~C</scope><scope>L~D</scope><scope>DOA</scope><orcidid>https://orcid.org/0000-0002-8787-6982</orcidid><orcidid>https://orcid.org/0000-0003-2192-2680</orcidid><orcidid>https://orcid.org/0000-0002-4321-379X</orcidid></search><sort><creationdate>2022</creationdate><title>Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech</title><author>Choi, Yeunju ; Jung, Youngmoon ; Suh, Youngjoo ; Kim, Hoirin</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c474t-e6e294493123a569b0e2b3a17bfb84577172a3ad0c6102b4cea7af63282e5c33</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2022</creationdate><topic>Data models</topic><topic>Distillation</topic><topic>Intelligibility</topic><topic>MOS prediction</topic><topic>neural TTS</topic><topic>perceptual loss</topic><topic>Prediction models</topic><topic>Predictions</topic><topic>Predictive models</topic><topic>Speech</topic><topic>Speech recognition</topic><topic>Speech synthesis</topic><topic>Task analysis</topic><topic>Training</topic><topic>Training data</topic><topic>Transformers</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Choi, Yeunju</creatorcontrib><creatorcontrib>Jung, Youngmoon</creatorcontrib><creatorcontrib>Suh, Youngjoo</creatorcontrib><creatorcontrib>Kim, Hoirin</creatorcontrib><collection>IEEE All-Society Periodicals Package (ASPP) 2005-present</collection><collection>IEEE Open Access Journals</collection><collection>IEEE All-Society Periodicals Package (ASPP) 1998-Present</collection><collection>IEEE Electronic Library (IEL)</collection><collection>CrossRef</collection><collection>Computer and Information Systems Abstracts</collection><collection>Electronics &amp; Communications Abstracts</collection><collection>Engineered Materials Abstracts</collection><collection>METADEX</collection><collection>Technology Research Database</collection><collection>Materials Research Database</collection><collection>ProQuest Computer Science Collection</collection><collection>Advanced Technologies Database with Aerospace</collection><collection>Computer and Information Systems Abstracts – Academic</collection><collection>Computer and Information Systems Abstracts Professional</collection><collection>DOAJ Directory of Open Access Journals</collection><jtitle>IEEE access</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Choi, Yeunju</au><au>Jung, Youngmoon</au><au>Suh, Youngjoo</au><au>Kim, Hoirin</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech</atitle><jtitle>IEEE access</jtitle><stitle>Access</stitle><date>2022</date><risdate>2022</risdate><volume>10</volume><spage>52621</spage><epage>52629</epage><pages>52621-52629</pages><issn>2169-3536</issn><eissn>2169-3536</eissn><coden>IAECCG</coden><abstract>Although recent neural text-to-speech (TTS) systems have achieved high-quality speech synthesis, there are cases where a TTS system generates low-quality speech, mainly caused by limited training data or information loss during knowledge distillation. Therefore, we propose a novel method to improve speech quality by training a TTS model under the supervision of perceptual loss, which measures the distance between the maximum possible speech quality score and the predicted one. We first pre-train a mean opinion score (MOS) prediction model and then train a TTS model to maximize the MOS of synthesized speech using the pre-trained MOS prediction model. The proposed method can be applied independently regardless of the TTS model architecture or the cause of speech quality degradation and efficiently without increasing the inference time or model complexity. The evaluation results for the MOS and phone error rate demonstrate that our proposed approach improves previous models in terms of both naturalness and intelligibility.</abstract><cop>Piscataway</cop><pub>IEEE</pub><doi>10.1109/ACCESS.2022.3175810</doi><tpages>9</tpages><orcidid>https://orcid.org/0000-0002-8787-6982</orcidid><orcidid>https://orcid.org/0000-0003-2192-2680</orcidid><orcidid>https://orcid.org/0000-0002-4321-379X</orcidid><oa>free_for_read</oa></addata></record>
fulltext fulltext
identifier ISSN: 2169-3536
ispartof IEEE access, 2022, Vol.10, p.52621-52629
issn 2169-3536
2169-3536
language eng
recordid cdi_ieee_primary_9775804
source IEEE Open Access Journals; DOAJ Directory of Open Access Journals; Elektronische Zeitschriftenbibliothek - Frei zugängliche E-Journals
subjects Data models
Distillation
Intelligibility
MOS prediction
neural TTS
perceptual loss
Prediction models
Predictions
Predictive models
Speech
Speech recognition
Speech synthesis
Task analysis
Training
Training data
Transformers
title Learning to Maximize Speech Quality Directly Using MOS Prediction for Neural Text-to-Speech
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-06T13%3A02%3A37IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_ieee_&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Learning%20to%20Maximize%20Speech%20Quality%20Directly%20Using%20MOS%20Prediction%20for%20Neural%20Text-to-Speech&rft.jtitle=IEEE%20access&rft.au=Choi,%20Yeunju&rft.date=2022&rft.volume=10&rft.spage=52621&rft.epage=52629&rft.pages=52621-52629&rft.issn=2169-3536&rft.eissn=2169-3536&rft.coden=IAECCG&rft_id=info:doi/10.1109/ACCESS.2022.3175810&rft_dat=%3Cproquest_ieee_%3E2669159184%3C/proquest_ieee_%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2669159184&rft_id=info:pmid/&rft_ieee_id=9775804&rft_doaj_id=oai_doaj_org_article_94c605add4db4b9fa8fb5f63361d4929&rfr_iscdi=true