The usage of multiple Hidden Markov Models (HMMs) to prepare a Czech acoustic unit inventory and speech synthesis based on this inventory are presented in this paper ( Triphone HMMs are trained on the basis of the speech corpus spoken by a single speaker.
)LJXUH SURGXFWLRQ 7KH FODVVLFDO DXWRUHJUHVVLYH PRGHO RI VSHHFK $5 DV\QFKURQRXV PRGHO RI VSHHFK SURGXFWLRQ VHH )LJXUH 3LWFKWKH /3 DQDO\VLV LV GRQH IRU HDFK VHJPHQW RIVSHHFK VSHHFK
3$5&25 VHJPHQW VHJPHQW GDWDEDVH $V D UHVXOW SDUDPHWULFSDUDPHWHUV FRHIILFLHQWV GDWDEDVH ZLWK LV RUGHU FRQVWUXFWHG :H XVH 9 89 IURP WKH QRQ SDUDPHWULF VSHHFK VHJPHQWV 9RFDO WUDFW LVGHFLVLRQ L H DQG GXUDWLRQ SLWFK LQWHQVLW\
FDQ EH FDOFXODWHG YRLFHG XQYRLFHG2WKHU V\QWKHVLV
GLUHFWO\
PRGHOHG 3$5&25V E\
LPSXOVH WUDLQ JHQHUDWRU LQ FDVH RI YRLFHG VRXQGV RU E\7KH D ODWWLFH H[FLWDWLRQ ILOWHU VLJQDO ZKRVH LV JHQHUDWHG FRHIILFLHQWV E\ DUHDQ*DXVVLDQ VWUDLJKWIRUZDUG QRLVH
H[SOLFLW SDUDPHWHUV RI WKH PRGHO 6SHFWUDO GLVFRQWLQXLWLHVVLQFH VRXUFH SLWFK GXUDWLRQ 3URVRG\ DQG PDWFKLQJ LQWHQVLW\ DUHLVHQFRXQWHUHG VPRRWK LQ WKH FRQFDWHQDWLRQ SRLQWV DUH VLPSOH WR3$5&25V E\ OLQHDU
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
PDUNV 7KHVH VKRUW WLPH VHJPHQWV DUH WKHQ FRPELQHG WRZLQGRZLQJ SLWFK V\QFKURQRXV FHQWHUHG VKRUW WLPH RQ WKH DQDO\VLV VHJPHQWV SLWFKE\SURGXFH D V\QWKHWLF VLJQDO 3URVRGLF PDWFKLQJ LV GRQH E\FKDQJLQJ DGGLQJ GLVWDQFHV EHWZHHQ
VLPSOH YDOXHV ZD\ RU RYHUODSSLQJ SLWFK DQG GXUDWLRQ VKRUW WLPH SLWFK SHULRG SXOVHV DQG E\
FDQ VHJPHQWV EH ,Q WKLV
VSHHFK VRPH LV :KLOH YHU\ VHJPHQWDO KLJK QR VSHHFK TXDOLW\ PRGHO RI WKH VHW WR GHVLUHG
LV V\QWKHVL]HG
SRLQWV SUREOHPV FDQ EH REVHUYHG LQ WKH FRQFDWHQDWLRQHPSOR\HG DSSOLHG 1R
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³)URP7H[W WR 6SHHFK 7KH 0,7DON 6\VWHP´ &DPEULGJH> @8QLYHUVLW\ 3UHVV &DPEULGJH 2OLYH - 3 9DQ 6DQWHQ - 0|ELXV % DQG 6KLQ & 6SHHFK 6\QWKHVLV 7KH %HOO /DEV $SSURDFKÄ6\QWKHVLV³ ,Q 0XOWLOLQJXDO 7H[W WR E\ +LUVFKEHUJ 9DQ 6DQWHQ - 6SURDW 5 : 2OLYH - 3 HGLWHGDQG> @'RUGUHFKW VWU - .OXZHU $FDGHPLF 3XEOLVKHUV 'RQRYDQ 6\QWKHVLV´ 5 ( > @&DPEULGJH 3K' WKHVLV 8QLYHUVLW\ RI &DPEULGJH ³$ 7UDLQDEOH 6SHHFK'RQRYDQ 7UDLQDEOH 6SHHFK 6\QWKHVLV 6\VWHP´ 5 ( (LGH 0 ³7KH ,%0> @RI ,&6/3¶ 3URFHHGLQJV+XDQJ *ROGVPLWK ; 6\GQH\ ³:KLVWOHU - $FHUR /LX $ $GFRFN - +RQ + : 3URFHHGLQJV RI ,&6/3¶ $ 7UDLQDEOH - DQG 7H[W WR 6SHHFK 3OXPSH 0 6\VWHP´ > @ 3KLODGHOSKLD SS <RXQJ :RRGODQG 6 -DQVHQ - 2GHOO - 2OODVRQ ' DQG> @&DPEULGJH 8QLYHUVLW\ &DPEULGJH 3 ³7KH +7. %RRN´ 'XWRLW 6SHHFK 7
'RUGUHFKW

6\QWKHVLV´ .OXZHU ³$Q ,QWURGXFWLRQ $FDGHPLF WR 3XEOLVKHUV 7H[W WR
Adaptation of pitch and spectrum for HMM-based speech synthesis using MLLR...based TTS system in which each speech synthesis unit is modeled by HMM [...
Corpus-based approaches to speech synthesis have been advocated to overcome the limitations of concatenative synthesis from a xed acoustic unit inventory. ...
USING HMM-BASED SPEECH RECOGNITION/SYNTHESIS ...cient vectors for each speech unit is modeled by...
Speech Technologies Acoustic Modeling of Speaking ...synthetic speech using HMM-based speech synthesis....speech parameters of a speech unit such as the ...
automatically, using HMM-based approach [12, 13]....: Join Cost for Unit Selection Speech Synthesis...Kavai, H., Tsuzaki, M.: Acoustic Measures vs....
Speech Synthesis Using H... 5页 免费 An HMM... HMM-BASED SMOOTHING FOR CONCATENATIVE SPEECH ...acoustic quality of the Whistler Text-to-Speech ...
An HMM-based speech synt... 暂无评价 4页 免费 HMM-Based Speech Synthes... 185页 免费 Speech Synthesis Using H... 5页 免费 HMM-BASED SMOOTHING ...
RULE-BASED VISUAL SPEECH SYNTHESIS Jonas Beskow ...speech, especially under degraded acoustic conditions...modalities to be written using the same notation....
called HMM-based generation synthesis (or simply...All of these techniques depend on a acoustic ...Ho?man, “Speech synthesis using stochastic Markov...
HMM-Based Speech Synthes... 185页 免费 Speech Synthesis Using H... 5...(for which the acoustic evaluation is based on an Euclidean distance measure...
热门文档