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.

Primena retke reprezentacije na modelima Gausovih mešavina koji se koriste za automatsko prepoznavanje govora

An application of sparse representation in Gaussian mixture models used inspeech recognition task

Thumbnail
2014
IzvestajKomisije2668.pdf (269.4Kb)
Disertacija2668.pdf (1.509Mb)
Author
Jakovljević, Nikša
Mentor
Delić, Vlado
Committee members
Trpovski, Željen
Jovičić, Slobodan
Grbić, Tatjana
Sečujski, Milan
Metadata
Show full item record
Abstract
U ovoj disertaciji je predstavljen model koji aproksimira pune kova- rijansne matrice u modelu gausovih mešavina (GMM) sa smanjenim brojem parametara i izračunavanja koji su potrebni za izračunavanje izglednosti. U predloženom modelu inverzne kovarijansne matrice su aproksimirane korišćenjem retke reprezentacije njihovih karakteri- stičnih vektora. Pored samog modela prikazan je i algoritam za estimaciju parametara zasnovan na kriterijumu maksimizacije izgeldnosti. Eksperimentalni rezultati na problemu prepoznavanja govora su pokazali da predloženi model za isti nivo greške kao GMM sa upunim kovarijansnim, redukuje broj parametara za 45%.
This thesis proposes a model which approximates full covariance matrices in Gaussian mixture models with a reduced number of parameters and computations required for likelihood evaluations. In the proposed model inverse covariance (precision) matrices are approximated using sparsely represented eigenvectors. A maximum likelihood algorithm for parameter estimation and its practical implementation are presented. Experimental results on a speech recognition task show that while keeping the word error rate close to the one obtained by GMMs with full covariance matrices, the proposed model can reduce the number of parameters by 45%.
Faculty:
Универзитет у Новом Саду, Факултет техничких наука
Date:
10-03-2014
Projects:
  • Development of Dialogue Systems for Serbian and Other South Slavic Languages (RS-32035)
Keywords:
prepoznavanje govora / Speech recognition / modeli Gausovih mešavina / retkareprezentacija / Gaussian mixture models / sparse representation

DOI: 10.2298/ns20131218jakovljevic

[ Google Scholar ]
Handle
https://hdl.handle.net/21.15107/rcub_nardus_5385
URI
https://nardus.mpn.gov.rs/handle/123456789/5385
http://www.cris.uns.ac.rs/DownloadFileServlet/Disertacija139566073394828.pdf?controlNumber=(BISIS)83705&fileName=139566073394828.pdf&id=1625&source=NaRDuS&language=sr
http://www.cris.uns.ac.rs/record.jsf?recordId=83705&source=NaRDuS&language=sr
http://www.cris.uns.ac.rs/DownloadFileServlet/IzvestajKomisijenobrazac6_Niksa_Jakovljevic.pdf?controlNumber=(BISIS)83705&fileName=nobrazac6_Niksa_Jakovljevic.pdf&id=932&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