Voice activity detection using audio-visual information

Theodore Petsatodis; Aristodemos Pnevmatikakis; Christos Boukis

doi:10.1109/ICDSP.2009.5201171

Voice activity detection using audio-visual information

Theodore Petsatodis
, Aristodemos Pnevmatikakis
, Christos Boukis

Department of Electronic Systems

Research output: Contribution to book/anthology/report/conference proceeding › Article in proceeding › Research › peer-review

14 Citations (Scopus)

Abstract

An audio-visual voice activity detector that uses sensors positioned distantly from the speaker is presented. Its constituting unimodal detectors are based on the modeling of the temporal variation of audio and visual features using Hidden Markov Models; their outcomes are fused using a post-decision scheme. The Mel-Frequency Cepstral Coefficients and the vertical mouth opening are the chosen audio and visual features respectively, both augmented with their first-order derivatives. The proposed system is assessed using far-field recordings from four different speakers and under various levels of additive white Gaussian noise, to obtain a performance superior than that which each unimodal component alone can achieve.

Original language	English
Title of host publication	DSP 2009: 16th International Conference on Digital Signal Processing, Proceedings
Number of pages	5
Publisher	IEEE Press
Publication date	2009
Pages	1-5
ISBN (Print)	978-142443298-1, 978-1-4244-3297-4
DOIs	https://doi.org/10.1109/ICDSP.2009.5201171
Publication status	Published - 2009
Event	DSP 2009: 16th International Conference on Digital Signal Processing - Santorini, Greece Duration: 5 Jul 2009 → 7 Jul 2009

Conference

Conference	DSP 2009: 16th International Conference on Digital Signal Processing
Country/Territory	Greece
City	Santorini
Period	05/07/2009 → 07/07/2009

Access to Document

10.1109/ICDSP.2009.5201171

AUB Link

Search for the material in Aalborg University Library's search engine

Cite this

@inproceedings{dd3cddde62504a5a83cbdb62031809ea,

title = "Voice activity detection using audio-visual information",

abstract = "An audio-visual voice activity detector that uses sensors positioned distantly from the speaker is presented. Its constituting unimodal detectors are based on the modeling of the temporal variation of audio and visual features using Hidden Markov Models; their outcomes are fused using a post-decision scheme. The Mel-Frequency Cepstral Coefficients and the vertical mouth opening are the chosen audio and visual features respectively, both augmented with their first-order derivatives. The proposed system is assessed using far-field recordings from four different speakers and under various levels of additive white Gaussian noise, to obtain a performance superior than that which each unimodal component alone can achieve.",

author = "Theodore Petsatodis and Aristodemos Pnevmatikakis and Christos Boukis",

note = "Article number 5201171; DSP 2009: 16th International Conference on Digital Signal Processing ; Conference date: 05-07-2009 Through 07-07-2009",

year = "2009",

doi = "10.1109/ICDSP.2009.5201171",

language = "English",

isbn = "978-142443298-1",

pages = "1--5",

booktitle = "DSP 2009: 16th International Conference on Digital Signal Processing, Proceedings",

publisher = "IEEE Press",

}

TY - GEN

T1 - Voice activity detection using audio-visual information

AU - Petsatodis, Theodore

AU - Pnevmatikakis, Aristodemos

AU - Boukis, Christos

N1 - Article number 5201171

PY - 2009

Y1 - 2009

N2 - An audio-visual voice activity detector that uses sensors positioned distantly from the speaker is presented. Its constituting unimodal detectors are based on the modeling of the temporal variation of audio and visual features using Hidden Markov Models; their outcomes are fused using a post-decision scheme. The Mel-Frequency Cepstral Coefficients and the vertical mouth opening are the chosen audio and visual features respectively, both augmented with their first-order derivatives. The proposed system is assessed using far-field recordings from four different speakers and under various levels of additive white Gaussian noise, to obtain a performance superior than that which each unimodal component alone can achieve.

AB - An audio-visual voice activity detector that uses sensors positioned distantly from the speaker is presented. Its constituting unimodal detectors are based on the modeling of the temporal variation of audio and visual features using Hidden Markov Models; their outcomes are fused using a post-decision scheme. The Mel-Frequency Cepstral Coefficients and the vertical mouth opening are the chosen audio and visual features respectively, both augmented with their first-order derivatives. The proposed system is assessed using far-field recordings from four different speakers and under various levels of additive white Gaussian noise, to obtain a performance superior than that which each unimodal component alone can achieve.

UR - https://www.scopus.com/pages/publications/70449555522

U2 - 10.1109/ICDSP.2009.5201171

DO - 10.1109/ICDSP.2009.5201171

M3 - Article in proceeding

AN - SCOPUS:70449555522

SN - 978-142443298-1

SN - 978-1-4244-3297-4

SP - 1

EP - 5

BT - DSP 2009: 16th International Conference on Digital Signal Processing, Proceedings

PB - IEEE Press

T2 - DSP 2009: 16th International Conference on Digital Signal Processing

Y2 - 5 July 2009 through 7 July 2009

ER -

Voice activity detection using audio-visual information

Abstract

Conference

Access to Document

AUB Link

Other files and links

Fingerprint

Cite this