A variational EM method for pole-zero modeling of speech with mixed block sparse and Gaussian excitation

Publikation: Bidrag til bog/antologi/rapport/konference proceedingKonferenceartikel i proceedingForskningpeer review

12 Downloads (Pure)

Resumé

The modeling of speech can be used for speech synthesis and speech recognition. We present a speech analysis method based on pole-zero modeling of speech with mixed block sparse and Gaussian excitation. By using a pole-zero model, instead of the all-pole model, a better spectral fitting can be expected. Moreover, motivated by the block sparse glottal flow excitation during voiced speech and the white noise excitation for unvoiced speech, we model the excitation sequence as a combination of block sparse signals and white noise. A variational EM (VEM) method is proposed for estimating the posterior PDFs of the block sparse residuals and point estimates of mod- elling parameters within a sparse Bayesian learning framework. Compared to conventional pole-zero and all-pole based methods, experimental results show that the proposed method has lower spectral distortion and good performance in reconstructing of the block sparse excitation.
OriginalsprogEngelsk
TitelThe 25th European Signal Processing Conference (EUSIPCO 2017)
Antal sider5
ForlagIEEE
Publikationsdato28 aug. 2017
Sider1784-1788
ISBN (Trykt)978-0-9928626-7-1
DOI
StatusUdgivet - 28 aug. 2017
Begivenhed25th European Signal Processing Conference 2017 - Kos International Convention Center, Kos, Grækenland
Varighed: 28 aug. 20172 sep. 2017
Konferencens nummer: 25
https://www.eusipco2017.org/#

Konference

Konference25th European Signal Processing Conference 2017
Nummer25
LokationKos International Convention Center
LandGrækenland
ByKos
Periode28/08/201702/09/2017
Internetadresse
NavnProceedings of the European Signal Processing Conference
ISSN2076-1465

Fingerprint

Poles
White noise
Poles and zeros
Speech synthesis
Speech analysis
Speech recognition

Citer dette

Shi, L., Nielsen, J. K., Jensen, J. R., & Christensen, M. G. (2017). A variational EM method for pole-zero modeling of speech with mixed block sparse and Gaussian excitation. I The 25th European Signal Processing Conference (EUSIPCO 2017) (s. 1784-1788). IEEE. Proceedings of the European Signal Processing Conference https://doi.org/10.23919/EUSIPCO.2017.8081516
Shi, Liming ; Nielsen, Jesper Kjær ; Jensen, Jesper Rindom ; Christensen, Mads Græsbøll. / A variational EM method for pole-zero modeling of speech with mixed block sparse and Gaussian excitation. The 25th European Signal Processing Conference (EUSIPCO 2017) . IEEE, 2017. s. 1784-1788 (Proceedings of the European Signal Processing Conference).
@inproceedings{0f5ea46107fa4fcb8de38f56bc782ed0,
title = "A variational EM method for pole-zero modeling of speech with mixed block sparse and Gaussian excitation",
abstract = "The modeling of speech can be used for speech synthesis and speech recognition. We present a speech analysis method based on pole-zero modeling of speech with mixed block sparse and Gaussian excitation. By using a pole-zero model, instead of the all-pole model, a better spectral fitting can be expected. Moreover, motivated by the block sparse glottal flow excitation during voiced speech and the white noise excitation for unvoiced speech, we model the excitation sequence as a combination of block sparse signals and white noise. A variational EM (VEM) method is proposed for estimating the posterior PDFs of the block sparse residuals and point estimates of mod- elling parameters within a sparse Bayesian learning framework. Compared to conventional pole-zero and all-pole based methods, experimental results show that the proposed method has lower spectral distortion and good performance in reconstructing of the block sparse excitation.",
author = "Liming Shi and Nielsen, {Jesper Kj{\ae}r} and Jensen, {Jesper Rindom} and Christensen, {Mads Gr{\ae}sb{\o}ll}",
year = "2017",
month = "8",
day = "28",
doi = "10.23919/EUSIPCO.2017.8081516",
language = "English",
isbn = "978-0-9928626-7-1",
pages = "1784--1788",
booktitle = "The 25th European Signal Processing Conference (EUSIPCO 2017)",
publisher = "IEEE",
address = "United States",

}

Shi, L, Nielsen, JK, Jensen, JR & Christensen, MG 2017, A variational EM method for pole-zero modeling of speech with mixed block sparse and Gaussian excitation. i The 25th European Signal Processing Conference (EUSIPCO 2017) . IEEE, Proceedings of the European Signal Processing Conference, s. 1784-1788, Kos, Grækenland, 28/08/2017. https://doi.org/10.23919/EUSIPCO.2017.8081516

A variational EM method for pole-zero modeling of speech with mixed block sparse and Gaussian excitation. / Shi, Liming; Nielsen, Jesper Kjær; Jensen, Jesper Rindom; Christensen, Mads Græsbøll.

The 25th European Signal Processing Conference (EUSIPCO 2017) . IEEE, 2017. s. 1784-1788 (Proceedings of the European Signal Processing Conference).

Publikation: Bidrag til bog/antologi/rapport/konference proceedingKonferenceartikel i proceedingForskningpeer review

TY - GEN

T1 - A variational EM method for pole-zero modeling of speech with mixed block sparse and Gaussian excitation

AU - Shi, Liming

AU - Nielsen, Jesper Kjær

AU - Jensen, Jesper Rindom

AU - Christensen, Mads Græsbøll

PY - 2017/8/28

Y1 - 2017/8/28

N2 - The modeling of speech can be used for speech synthesis and speech recognition. We present a speech analysis method based on pole-zero modeling of speech with mixed block sparse and Gaussian excitation. By using a pole-zero model, instead of the all-pole model, a better spectral fitting can be expected. Moreover, motivated by the block sparse glottal flow excitation during voiced speech and the white noise excitation for unvoiced speech, we model the excitation sequence as a combination of block sparse signals and white noise. A variational EM (VEM) method is proposed for estimating the posterior PDFs of the block sparse residuals and point estimates of mod- elling parameters within a sparse Bayesian learning framework. Compared to conventional pole-zero and all-pole based methods, experimental results show that the proposed method has lower spectral distortion and good performance in reconstructing of the block sparse excitation.

AB - The modeling of speech can be used for speech synthesis and speech recognition. We present a speech analysis method based on pole-zero modeling of speech with mixed block sparse and Gaussian excitation. By using a pole-zero model, instead of the all-pole model, a better spectral fitting can be expected. Moreover, motivated by the block sparse glottal flow excitation during voiced speech and the white noise excitation for unvoiced speech, we model the excitation sequence as a combination of block sparse signals and white noise. A variational EM (VEM) method is proposed for estimating the posterior PDFs of the block sparse residuals and point estimates of mod- elling parameters within a sparse Bayesian learning framework. Compared to conventional pole-zero and all-pole based methods, experimental results show that the proposed method has lower spectral distortion and good performance in reconstructing of the block sparse excitation.

U2 - 10.23919/EUSIPCO.2017.8081516

DO - 10.23919/EUSIPCO.2017.8081516

M3 - Article in proceeding

SN - 978-0-9928626-7-1

SP - 1784

EP - 1788

BT - The 25th European Signal Processing Conference (EUSIPCO 2017)

PB - IEEE

ER -

Shi L, Nielsen JK, Jensen JR, Christensen MG. A variational EM method for pole-zero modeling of speech with mixed block sparse and Gaussian excitation. I The 25th European Signal Processing Conference (EUSIPCO 2017) . IEEE. 2017. s. 1784-1788. (Proceedings of the European Signal Processing Conference). https://doi.org/10.23919/EUSIPCO.2017.8081516