National Repository of Dissertations in Serbia
    • English
    • Српски
    • Српски (Serbia)
  • English 
    • English
    • Serbian (Cyrilic)
    • Serbian (Latin)
  • Login
View Item 
  •   NaRDuS home
  • Универзитет у Новом Саду
  • Факултет техничких наука
  • View Item
  •   NaRDuS home
  • Универзитет у Новом Саду
  • Факултет техничких наука
  • View Item
JavaScript is disabled for your browser. Some features of this site may not work without it.

Parametarska sinteza ekspresivnog govora

Parametric synthesis of expressive speech

Thumbnail
2019
Disertacija.pdf (2.391Mb)
IzvestajKomisije.pdf (448.8Kb)
Author
Suzić, Siniša
Mentor
Delić, Vlado
Committee members
Sečujski, Milan
Trpovski, Željen
Grbić, Tatjana
Perić, Zoran
Jakovljević, Nikša
Delić, Vlado
Metadata
Show full item record
Abstract
U disertaciji su opisani postupci sinteze ekspresivnog govora korišćenjem parametarskih pristupa. Pokazano je da se korišćenjem dubokih neuronskih mreža dobijaju bolji rezultati nego korišćenjem skrivenix Markovljevih modela. Predložene su tri nove metode za sintezu ekspresivnog govora korišćenjem dubokih neuronskih mreža: metoda kodova stila, metoda dodatne obuke mreže i arhitektura zasnovana na deljenim skrivenim slojevima. Pokazano je da se najbolji rezultati dobijaju korišćenjem metode kodova stila. Takođe je predložana i nova metoda za transplantaciju emocija/stilova bazirana na deljenim skrivenim slojevima. Predložena metoda ocenjena je bolje od referentne metode iz literature.
In this thesis methods for expressive speech synthesis using parametric approaches are presented. It is shown that better results are achived with usage of deep neural networks compared to synthesis based on hidden Markov models. Three new methods for synthesis of expresive speech using deep neural networks are presented: style codes, model re-training and shared hidden layer architecture. It is shown that best results are achived by using style code method. The new method for style transplantation based on shared hidden layer architecture is also proposed. It is shown that this method outperforms referent method from literature.
Faculty:
Универзитет у Новом Саду, Факултет техничких наука
Date:
12-07-2019
Keywords:
sinteza govora / text-to-speech synthesis / ekspresivni govor / neuronske mreže / skriveniMarkovljevi modeli / expressive speech / neural networks / hidden Markov models
[ Google Scholar ]
Handle
https://hdl.handle.net/21.15107/rcub_nardus_11230
URI
https://www.cris.uns.ac.rs/DownloadFileServlet/Disertacija155609191109585.pdf?controlNumber=(BISIS)110631&fileName=155609191109585.pdf&id=12807&source=NaRDuS&language=sr
https://nardus.mpn.gov.rs/handle/123456789/11230
https://www.cris.uns.ac.rs/record.jsf?recordId=110631&source=NaRDuS&language=sr
https://www.cris.uns.ac.rs/DownloadFileServlet/IzvestajKomisije155609192458893.pdf?controlNumber=(BISIS)110631&fileName=155609192458893.pdf&id=12808&source=NaRDuS&language=sr

DSpace software copyright © 2002-2015  DuraSpace
About NaRDus | Contact us

OpenAIRERCUBRODOSTEMPUS
 

 

Browse

All of DSpaceUniversities & FacultiesAuthorsMentorCommittee membersSubjectsThis CollectionAuthorsMentorCommittee membersSubjects

DSpace software copyright © 2002-2015  DuraSpace
About NaRDus | Contact us

OpenAIRERCUBRODOSTEMPUS