SAR: Self-Supervised Anti-Distortion Representation for End-To-End Speech Model

In recent Text-to-Speech (TTS) systems, a neural vocoder often generates speech samples by solely conditioning on acoustic features predicted from an acoustic model. However, there are always distortions existing in the predicted acoustic features, compared to those of the groundtruth, especially in...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wang, Jianzong, Zhang, Xulong, Tang, Haobin, Sun, Aolan, Cheng, Ning, Xiao, Jing
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Wang, Jianzong
Zhang, Xulong
Tang, Haobin
Sun, Aolan
Cheng, Ning
Xiao, Jing
description In recent Text-to-Speech (TTS) systems, a neural vocoder often generates speech samples by solely conditioning on acoustic features predicted from an acoustic model. However, there are always distortions existing in the predicted acoustic features, compared to those of the groundtruth, especially in the common case of poor acoustic modeling due to low-quality training data. To overcome such limits, we propose a Self-supervised learning framework to learn an Anti-distortion acoustic Representation (SAR) to replace human-crafted acoustic features by introducing distortion prior to an auto-encoder pre-training process. The learned acoustic representation from the proposed framework is proved anti-distortion compared to the most commonly used mel-spectrogram through both objective and subjective evaluation.
doi_str_mv 10.48550/arxiv.2304.11547
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2304_11547</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2304_11547</sourcerecordid><originalsourceid>FETCH-LOGICAL-a677-a35c1a7b4111f301b15fbf610c329ff4948d1185a43567cac27349f698b53c853</originalsourceid><addsrcrecordid>eNotz71OwzAYhWEvDKjlApjwDTj4i-3Y6RaV8iMVVWqyR47zWVgKceSECu4eCEyvznKkh5Bb4Jk0SvF7mz7DJcsFlxmAkvqanOrqvKM1Dp7VHxOmS5ixp9W4BPYQ5iWmJcSRnnFKOOO42HX6mOhh7FkT2U9oPSG6N_oaexy25MrbYcab_25I83ho9s_seHp62VdHZgutmRXKgdWdBAAvOHSgfOcL4E7kpfeylKYHMMpKoQrtrMu1kKUvStMp4YwSG3L3d7uK2imFd5u-2l9Zu8rENyHbR4M</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>SAR: Self-Supervised Anti-Distortion Representation for End-To-End Speech Model</title><source>arXiv.org</source><creator>Wang, Jianzong ; Zhang, Xulong ; Tang, Haobin ; Sun, Aolan ; Cheng, Ning ; Xiao, Jing</creator><creatorcontrib>Wang, Jianzong ; Zhang, Xulong ; Tang, Haobin ; Sun, Aolan ; Cheng, Ning ; Xiao, Jing</creatorcontrib><description>In recent Text-to-Speech (TTS) systems, a neural vocoder often generates speech samples by solely conditioning on acoustic features predicted from an acoustic model. However, there are always distortions existing in the predicted acoustic features, compared to those of the groundtruth, especially in the common case of poor acoustic modeling due to low-quality training data. To overcome such limits, we propose a Self-supervised learning framework to learn an Anti-distortion acoustic Representation (SAR) to replace human-crafted acoustic features by introducing distortion prior to an auto-encoder pre-training process. The learned acoustic representation from the proposed framework is proved anti-distortion compared to the most commonly used mel-spectrogram through both objective and subjective evaluation.</description><identifier>DOI: 10.48550/arxiv.2304.11547</identifier><language>eng</language><subject>Computer Science - Sound</subject><creationdate>2023-04</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,780,885</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2304.11547$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2304.11547$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Wang, Jianzong</creatorcontrib><creatorcontrib>Zhang, Xulong</creatorcontrib><creatorcontrib>Tang, Haobin</creatorcontrib><creatorcontrib>Sun, Aolan</creatorcontrib><creatorcontrib>Cheng, Ning</creatorcontrib><creatorcontrib>Xiao, Jing</creatorcontrib><title>SAR: Self-Supervised Anti-Distortion Representation for End-To-End Speech Model</title><description>In recent Text-to-Speech (TTS) systems, a neural vocoder often generates speech samples by solely conditioning on acoustic features predicted from an acoustic model. However, there are always distortions existing in the predicted acoustic features, compared to those of the groundtruth, especially in the common case of poor acoustic modeling due to low-quality training data. To overcome such limits, we propose a Self-supervised learning framework to learn an Anti-distortion acoustic Representation (SAR) to replace human-crafted acoustic features by introducing distortion prior to an auto-encoder pre-training process. The learned acoustic representation from the proposed framework is proved anti-distortion compared to the most commonly used mel-spectrogram through both objective and subjective evaluation.</description><subject>Computer Science - Sound</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotz71OwzAYhWEvDKjlApjwDTj4i-3Y6RaV8iMVVWqyR47zWVgKceSECu4eCEyvznKkh5Bb4Jk0SvF7mz7DJcsFlxmAkvqanOrqvKM1Dp7VHxOmS5ixp9W4BPYQ5iWmJcSRnnFKOOO42HX6mOhh7FkT2U9oPSG6N_oaexy25MrbYcab_25I83ho9s_seHp62VdHZgutmRXKgdWdBAAvOHSgfOcL4E7kpfeylKYHMMpKoQrtrMu1kKUvStMp4YwSG3L3d7uK2imFd5u-2l9Zu8rENyHbR4M</recordid><startdate>20230423</startdate><enddate>20230423</enddate><creator>Wang, Jianzong</creator><creator>Zhang, Xulong</creator><creator>Tang, Haobin</creator><creator>Sun, Aolan</creator><creator>Cheng, Ning</creator><creator>Xiao, Jing</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20230423</creationdate><title>SAR: Self-Supervised Anti-Distortion Representation for End-To-End Speech Model</title><author>Wang, Jianzong ; Zhang, Xulong ; Tang, Haobin ; Sun, Aolan ; Cheng, Ning ; Xiao, Jing</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a677-a35c1a7b4111f301b15fbf610c329ff4948d1185a43567cac27349f698b53c853</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Computer Science - Sound</topic><toplevel>online_resources</toplevel><creatorcontrib>Wang, Jianzong</creatorcontrib><creatorcontrib>Zhang, Xulong</creatorcontrib><creatorcontrib>Tang, Haobin</creatorcontrib><creatorcontrib>Sun, Aolan</creatorcontrib><creatorcontrib>Cheng, Ning</creatorcontrib><creatorcontrib>Xiao, Jing</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Wang, Jianzong</au><au>Zhang, Xulong</au><au>Tang, Haobin</au><au>Sun, Aolan</au><au>Cheng, Ning</au><au>Xiao, Jing</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>SAR: Self-Supervised Anti-Distortion Representation for End-To-End Speech Model</atitle><date>2023-04-23</date><risdate>2023</risdate><abstract>In recent Text-to-Speech (TTS) systems, a neural vocoder often generates speech samples by solely conditioning on acoustic features predicted from an acoustic model. However, there are always distortions existing in the predicted acoustic features, compared to those of the groundtruth, especially in the common case of poor acoustic modeling due to low-quality training data. To overcome such limits, we propose a Self-supervised learning framework to learn an Anti-distortion acoustic Representation (SAR) to replace human-crafted acoustic features by introducing distortion prior to an auto-encoder pre-training process. The learned acoustic representation from the proposed framework is proved anti-distortion compared to the most commonly used mel-spectrogram through both objective and subjective evaluation.</abstract><doi>10.48550/arxiv.2304.11547</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2304.11547
ispartof
issn
language eng
recordid cdi_arxiv_primary_2304_11547
source arXiv.org
subjects Computer Science - Sound
title SAR: Self-Supervised Anti-Distortion Representation for End-To-End Speech Model
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-30T21%3A42%3A30IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=SAR:%20Self-Supervised%20Anti-Distortion%20Representation%20for%20End-To-End%20Speech%20Model&rft.au=Wang,%20Jianzong&rft.date=2023-04-23&rft_id=info:doi/10.48550/arxiv.2304.11547&rft_dat=%3Carxiv_GOX%3E2304_11547%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true