Regularized linear prediction all-pole models

M. N. Murthi; W. B. Kleijn

doi:10.1109/SCFT.2000.878410

Regularized linear prediction all-pole models

M. N. Murthi, W. B. Kleijn

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

13 Scopus citations

Abstract

For many cases of voiced speech, linear prediction (LP) based all-pole spectral envelopes exhibit unnatural vocal tract transfer functions that underestimate the formant bandwidths. To obtain smoother contoured all-pole spectral envelopes, we employ a regularization measure which discourages nonsmooth behavior of the transfer function. In particular, we demonstrate how a simple regularization scheme can be incorporated into the LP framework without the need for iterative numerical optimization or spectral sampling. Our results indicate that regularized LP all-pole models can provide more accurate vocal tract transfer function modeling than conventional LP, particularly at the formants.

Original language	English (US)
Title of host publication	2000 IEEE Workshop on Speech Coding - Proceedings
Subtitle of host publication	Meeting the Challenges of the New Millennium
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	96-98
Number of pages	3
ISBN (Electronic)	0780364163, 9780780364165
DOIs	https://doi.org/10.1109/SCFT.2000.878410
State	Published - Jan 1 2000
Externally published	Yes
Event	IEEE Workshop on Speech Coding - Delavan, United States Duration: Sep 17 2000 → Sep 20 2000

Publication series

Name	2000 IEEE Workshop on Speech Coding - Proceedings: Meeting the Challenges of the New Millennium

Other

Other	IEEE Workshop on Speech Coding
Country	United States
City	Delavan
Period	9/17/00 → 9/20/00

Keywords

Bandwidth
Cost function
Frequency estimation
Parameter estimation
Power harmonic filters
Predictive models
Signal synthesis
Speech coding
Speech synthesis
Transfer functions

ASJC Scopus subject areas

Signal Processing
Software

Access to Document

10.1109/SCFT.2000.878410

Fingerprint Dive into the research topics of 'Regularized linear prediction all-pole models'. Together they form a unique fingerprint.

Cite this

Murthi, M. N., & Kleijn, W. B. (2000). Regularized linear prediction all-pole models. In 2000 IEEE Workshop on Speech Coding - Proceedings: Meeting the Challenges of the New Millennium (pp. 96-98). [878410] (2000 IEEE Workshop on Speech Coding - Proceedings: Meeting the Challenges of the New Millennium). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/SCFT.2000.878410

Regularized linear prediction all-pole models. / Murthi, M. N.; Kleijn, W. B.

2000 IEEE Workshop on Speech Coding - Proceedings: Meeting the Challenges of the New Millennium. Institute of Electrical and Electronics Engineers Inc., 2000. p. 96-98 878410 (2000 IEEE Workshop on Speech Coding - Proceedings: Meeting the Challenges of the New Millennium).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Murthi, MN & Kleijn, WB 2000, Regularized linear prediction all-pole models. in 2000 IEEE Workshop on Speech Coding - Proceedings: Meeting the Challenges of the New Millennium., 878410, 2000 IEEE Workshop on Speech Coding - Proceedings: Meeting the Challenges of the New Millennium, Institute of Electrical and Electronics Engineers Inc., pp. 96-98, IEEE Workshop on Speech Coding, Delavan, United States, 9/17/00. https://doi.org/10.1109/SCFT.2000.878410

Murthi MN, Kleijn WB. Regularized linear prediction all-pole models. In 2000 IEEE Workshop on Speech Coding - Proceedings: Meeting the Challenges of the New Millennium. Institute of Electrical and Electronics Engineers Inc. 2000. p. 96-98. 878410. (2000 IEEE Workshop on Speech Coding - Proceedings: Meeting the Challenges of the New Millennium). https://doi.org/10.1109/SCFT.2000.878410

@inproceedings{c48308d2578243d18e4be9bcb0bd16cd,

title = "Regularized linear prediction all-pole models",

abstract = "For many cases of voiced speech, linear prediction (LP) based all-pole spectral envelopes exhibit unnatural vocal tract transfer functions that underestimate the formant bandwidths. To obtain smoother contoured all-pole spectral envelopes, we employ a regularization measure which discourages nonsmooth behavior of the transfer function. In particular, we demonstrate how a simple regularization scheme can be incorporated into the LP framework without the need for iterative numerical optimization or spectral sampling. Our results indicate that regularized LP all-pole models can provide more accurate vocal tract transfer function modeling than conventional LP, particularly at the formants.",

keywords = "Bandwidth, Cost function, Frequency estimation, Parameter estimation, Power harmonic filters, Predictive models, Signal synthesis, Speech coding, Speech synthesis, Transfer functions",

author = "Murthi, {M. N.} and Kleijn, {W. B.}",

year = "2000",

month = jan,

day = "1",

doi = "10.1109/SCFT.2000.878410",

language = "English (US)",

series = "2000 IEEE Workshop on Speech Coding - Proceedings: Meeting the Challenges of the New Millennium",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "96--98",

booktitle = "2000 IEEE Workshop on Speech Coding - Proceedings",

note = "IEEE Workshop on Speech Coding ; Conference date: 17-09-2000 Through 20-09-2000",

}

TY - GEN

T1 - Regularized linear prediction all-pole models

AU - Murthi, M. N.

AU - Kleijn, W. B.

PY - 2000/1/1

Y1 - 2000/1/1

N2 - For many cases of voiced speech, linear prediction (LP) based all-pole spectral envelopes exhibit unnatural vocal tract transfer functions that underestimate the formant bandwidths. To obtain smoother contoured all-pole spectral envelopes, we employ a regularization measure which discourages nonsmooth behavior of the transfer function. In particular, we demonstrate how a simple regularization scheme can be incorporated into the LP framework without the need for iterative numerical optimization or spectral sampling. Our results indicate that regularized LP all-pole models can provide more accurate vocal tract transfer function modeling than conventional LP, particularly at the formants.

AB - For many cases of voiced speech, linear prediction (LP) based all-pole spectral envelopes exhibit unnatural vocal tract transfer functions that underestimate the formant bandwidths. To obtain smoother contoured all-pole spectral envelopes, we employ a regularization measure which discourages nonsmooth behavior of the transfer function. In particular, we demonstrate how a simple regularization scheme can be incorporated into the LP framework without the need for iterative numerical optimization or spectral sampling. Our results indicate that regularized LP all-pole models can provide more accurate vocal tract transfer function modeling than conventional LP, particularly at the formants.

KW - Bandwidth

KW - Cost function

KW - Frequency estimation

KW - Parameter estimation

KW - Power harmonic filters

KW - Predictive models

KW - Signal synthesis

KW - Speech coding

KW - Speech synthesis

KW - Transfer functions

UR - http://www.scopus.com/inward/record.url?scp=0037662531&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0037662531&partnerID=8YFLogxK

U2 - 10.1109/SCFT.2000.878410

DO - 10.1109/SCFT.2000.878410

M3 - Conference contribution

AN - SCOPUS:0037662531

T3 - 2000 IEEE Workshop on Speech Coding - Proceedings: Meeting the Challenges of the New Millennium

SP - 96

EP - 98

BT - 2000 IEEE Workshop on Speech Coding - Proceedings

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - IEEE Workshop on Speech Coding

Y2 - 17 September 2000 through 20 September 2000

ER -