Convex Combination of Multiple Statistical Models with Application to VAD

Theodoros Petsatodis, Christos Boukis, Fotios Talantzis, Zheng-Hua Tan, Ramjee Prasad

Research output: Contribution to journalJournal articleResearchpeer-review

21 Citations (Scopus)

Abstract

This paper proposes a robust Voice Activity Detector (VAD) based on the observation that the distribution of speech captured with far-field microphones is highly varying, depending on the noise and reverberation conditions. The proposed VAD employs a convex combination scheme comprising three statistical distributions - a Gaussian, a Laplacian, and a two-sided Gamma - to effectively model captured speech. This scheme shows increased ability to adapt to dynamic acoustic environments. The contribution of each distribution to this convex combination is automatically adjusted based on the statistical characteristics of the instantaneous audio input. To further improve the performance of the system, an adaptive threshold is introduced, while a decision-smoothing scheme caters to the intra-frame correlation of speech signals. Extensive experiments under realistic scenarios support the proposed approach of combining several models for increased adaptation and performance.
Original languageEnglish
JournalI E E E Transactions on Audio, Speech and Language Processing
Volume19
Issue number8
Pages (from-to)2314-2327
ISSN1558-7916
DOIs
Publication statusPublished - Nov 2011

Fingerprint

Detectors
detectors
reverberation
statistical distributions
microphones
smoothing
far fields
Reverberation
Adaptive systems
Microphones
thresholds
Acoustic noise
acoustics
Acoustics
Statistical Models
Experiments

Keywords

  • voice activity detection
  • convex combination
  • classification
  • statistical models

Cite this

Petsatodis, Theodoros ; Boukis, Christos ; Talantzis, Fotios ; Tan, Zheng-Hua ; Prasad, Ramjee. / Convex Combination of Multiple Statistical Models with Application to VAD. In: I E E E Transactions on Audio, Speech and Language Processing. 2011 ; Vol. 19, No. 8. pp. 2314-2327.
@article{7c38625254a7492d8e055d31ba630ddb,
title = "Convex Combination of Multiple Statistical Models with Application to VAD",
abstract = "This paper proposes a robust Voice Activity Detector (VAD) based on the observation that the distribution of speech captured with far-field microphones is highly varying, depending on the noise and reverberation conditions. The proposed VAD employs a convex combination scheme comprising three statistical distributions - a Gaussian, a Laplacian, and a two-sided Gamma - to effectively model captured speech. This scheme shows increased ability to adapt to dynamic acoustic environments. The contribution of each distribution to this convex combination is automatically adjusted based on the statistical characteristics of the instantaneous audio input. To further improve the performance of the system, an adaptive threshold is introduced, while a decision-smoothing scheme caters to the intra-frame correlation of speech signals. Extensive experiments under realistic scenarios support the proposed approach of combining several models for increased adaptation and performance.",
keywords = "voice activity detection, convex combination, classification , statistical models",
author = "Theodoros Petsatodis and Christos Boukis and Fotios Talantzis and Zheng-Hua Tan and Ramjee Prasad",
year = "2011",
month = "11",
doi = "10.1109/TASL.2011.2131131",
language = "English",
volume = "19",
pages = "2314--2327",
journal = "IEEE/ACM Transactions on Audio, Speech, and Language Processing",
issn = "2329-9290",
publisher = "IEEE Signal Processing Society",
number = "8",

}

Convex Combination of Multiple Statistical Models with Application to VAD. / Petsatodis, Theodoros; Boukis, Christos ; Talantzis, Fotios ; Tan, Zheng-Hua; Prasad, Ramjee.

In: I E E E Transactions on Audio, Speech and Language Processing, Vol. 19, No. 8, 11.2011, p. 2314-2327.

Research output: Contribution to journalJournal articleResearchpeer-review

TY - JOUR

T1 - Convex Combination of Multiple Statistical Models with Application to VAD

AU - Petsatodis, Theodoros

AU - Boukis, Christos

AU - Talantzis, Fotios

AU - Tan, Zheng-Hua

AU - Prasad, Ramjee

PY - 2011/11

Y1 - 2011/11

N2 - This paper proposes a robust Voice Activity Detector (VAD) based on the observation that the distribution of speech captured with far-field microphones is highly varying, depending on the noise and reverberation conditions. The proposed VAD employs a convex combination scheme comprising three statistical distributions - a Gaussian, a Laplacian, and a two-sided Gamma - to effectively model captured speech. This scheme shows increased ability to adapt to dynamic acoustic environments. The contribution of each distribution to this convex combination is automatically adjusted based on the statistical characteristics of the instantaneous audio input. To further improve the performance of the system, an adaptive threshold is introduced, while a decision-smoothing scheme caters to the intra-frame correlation of speech signals. Extensive experiments under realistic scenarios support the proposed approach of combining several models for increased adaptation and performance.

AB - This paper proposes a robust Voice Activity Detector (VAD) based on the observation that the distribution of speech captured with far-field microphones is highly varying, depending on the noise and reverberation conditions. The proposed VAD employs a convex combination scheme comprising three statistical distributions - a Gaussian, a Laplacian, and a two-sided Gamma - to effectively model captured speech. This scheme shows increased ability to adapt to dynamic acoustic environments. The contribution of each distribution to this convex combination is automatically adjusted based on the statistical characteristics of the instantaneous audio input. To further improve the performance of the system, an adaptive threshold is introduced, while a decision-smoothing scheme caters to the intra-frame correlation of speech signals. Extensive experiments under realistic scenarios support the proposed approach of combining several models for increased adaptation and performance.

KW - voice activity detection

KW - convex combination

KW - classification

KW - statistical models

U2 - 10.1109/TASL.2011.2131131

DO - 10.1109/TASL.2011.2131131

M3 - Journal article

VL - 19

SP - 2314

EP - 2327

JO - IEEE/ACM Transactions on Audio, Speech, and Language Processing

JF - IEEE/ACM Transactions on Audio, Speech, and Language Processing

SN - 2329-9290

IS - 8

ER -