HMM-BASED HIERARCHICALUNITSELECTIONCOMBINING KULLBACK-LEIBLER DIVERGENCE WITH LIKELIHOODCRITERION
Lists
Only you see this
Abstract
Thispaper presents ahidden Markovmodel(HMM)based unit selection methodusinghierarchical units understatistical criterion. Inourprevious workwetried touseframesized speech segments andmaximumlikelihood criterion toimprove theperformance oftraditional concatenative synthesis system using phonesized units andcostfunction criterion. Inthis paper, hierarchical units whichconsist ofphonelevel units and frame level units areadopted toachieve better balance between thecoverage rateofcandidate unitandthenumberof concatenation points during synthesis. Besides, KullbackLeibler divergence(KLD) betweencandidate andtarget phoneme HMMs isintroduced asapart ofthefinal criterion for unitselection. Thelistening result provesthatthesetwo approaches canimprove theperformance ofsynthetic speech effectively. IndexTerms-Speech Synthesis, HMM,KLD
BibTeX
@inproceedings{ling2007hmm,
author = {Zhen-Hua Ling},
title = {{HMM-BASED HIERARCHICALUNITSELECTIONCOMBINING KULLBACK-LEIBLER DIVERGENCE WITH LIKELIHOODCRITERION}},
booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
year = {2007},
url = {https://openreview.net/forum?id=ohTuo5xFa4},
}Loading related content...