Distributed Expectation-Maximization Algorithm for Speaker Localization in Reverberant Environments

Yuval Dorfan; Axel Plinge; Gershon Hazan; Sharon Gannot

doi:10.1109/TASLP.2017.2788198

Distributed Expectation-Maximization Algorithm for Speaker Localization in Reverberant Environments

Yuval Dorfan, Axel Plinge, Gershon Hazan, Sharon Gannot

Bar-Ilan University - The Alexander Kofkin Faculty of Engineering

Research output: Contribution to journal › Article › peer-review

17 Scopus citations

Abstract

Localization of acoustic sources has attracted a considerable amount of research attention in recent years. A major obstacle to achieving high localization accuracy is the presence of reverberation, the influence of which obviously increases with the number of active speakers in the room. Human hearing is capable of localizing acoustic sources even in extreme conditions. In this study, we propose to combine a method based on human hearing mechanisms and a modified incremental distributed expectation-maximization (IDEM) algorithm. Rather than using phase difference measurements that are modeled by a mixture of complex-valued Gaussians, as proposed in the original IDEM framework, we propose to use time difference of arrival measurements in multiple subbands and model them by a mixture of real-valued truncated Gaussians. Moreover, we propose to first filter the measurements in order to reduce the effect of the multipath conditions. The proposed method is evaluated using both simulated data and real-life recordings.

Original language	English
Article number	8241767
Pages (from-to)	682-695
Number of pages	14
Journal	IEEE/ACM Transactions on Audio Speech and Language Processing
Volume	26
Issue number	3
DOIs	https://doi.org/10.1109/TASLP.2017.2788198
State	Published - Mar 2018

Bibliographical note

Funding Information:
Manuscript received July 12, 2017; revised November 17, 2017; accepted December 18, 2017. Date of publication December 29, 2017; date of current version January 25, 2018. The work of A. Plinge was supported by a fellowship within the FITweltweit Programme of the German Academic Exchange Service (DAAD). The associate editor coordinating the review of this manuscript and approving it for publication was Prof. Mads Græsbøll Christensen. (Corresponding author: Sharon Gannot.) Y. Dorfan, G. Hazan, and S. Gannot are with the Faculty of Engineering, Bar-Ilan University, Ramat-Gan 5290002, Israel (e-mail: dorfany@gmail.com; hazanshl@gmail.com; Sharon.Gannot@biu.ac.il).

Publisher Copyright:
© 2014 IEEE.

Keywords

Precedence effect
auditory scene analysis
distributed expectation-maximization
incremental expectation-maximization
multi-path
onset dominance
sound source localization
spectral masking
time difference of arrival
truncated Gaussian

Access to Document

10.1109/TASLP.2017.2788198

Cite this

@article{f6951dd5131342548b7c42b5d44b5b5a,

title = "Distributed Expectation-Maximization Algorithm for Speaker Localization in Reverberant Environments",

abstract = "Localization of acoustic sources has attracted a considerable amount of research attention in recent years. A major obstacle to achieving high localization accuracy is the presence of reverberation, the influence of which obviously increases with the number of active speakers in the room. Human hearing is capable of localizing acoustic sources even in extreme conditions. In this study, we propose to combine a method based on human hearing mechanisms and a modified incremental distributed expectation-maximization (IDEM) algorithm. Rather than using phase difference measurements that are modeled by a mixture of complex-valued Gaussians, as proposed in the original IDEM framework, we propose to use time difference of arrival measurements in multiple subbands and model them by a mixture of real-valued truncated Gaussians. Moreover, we propose to first filter the measurements in order to reduce the effect of the multipath conditions. The proposed method is evaluated using both simulated data and real-life recordings.",

keywords = "Precedence effect, auditory scene analysis, distributed expectation-maximization, incremental expectation-maximization, multi-path, onset dominance, sound source localization, spectral masking, time difference of arrival, truncated Gaussian",

author = "Yuval Dorfan and Axel Plinge and Gershon Hazan and Sharon Gannot",

note = "Publisher Copyright: {\textcopyright} 2014 IEEE.",

year = "2018",

month = mar,

doi = "10.1109/TASLP.2017.2788198",

language = "אנגלית",

volume = "26",

pages = "682--695",

journal = "IEEE/ACM Transactions on Audio Speech and Language Processing",

issn = "2329-9290",

publisher = "IEEE Advancing Technology for Humanity",

number = "3",

}

TY - JOUR

T1 - Distributed Expectation-Maximization Algorithm for Speaker Localization in Reverberant Environments

AU - Dorfan, Yuval

AU - Plinge, Axel

AU - Hazan, Gershon

AU - Gannot, Sharon

PY - 2018/3

Y1 - 2018/3

N2 - Localization of acoustic sources has attracted a considerable amount of research attention in recent years. A major obstacle to achieving high localization accuracy is the presence of reverberation, the influence of which obviously increases with the number of active speakers in the room. Human hearing is capable of localizing acoustic sources even in extreme conditions. In this study, we propose to combine a method based on human hearing mechanisms and a modified incremental distributed expectation-maximization (IDEM) algorithm. Rather than using phase difference measurements that are modeled by a mixture of complex-valued Gaussians, as proposed in the original IDEM framework, we propose to use time difference of arrival measurements in multiple subbands and model them by a mixture of real-valued truncated Gaussians. Moreover, we propose to first filter the measurements in order to reduce the effect of the multipath conditions. The proposed method is evaluated using both simulated data and real-life recordings.

AB - Localization of acoustic sources has attracted a considerable amount of research attention in recent years. A major obstacle to achieving high localization accuracy is the presence of reverberation, the influence of which obviously increases with the number of active speakers in the room. Human hearing is capable of localizing acoustic sources even in extreme conditions. In this study, we propose to combine a method based on human hearing mechanisms and a modified incremental distributed expectation-maximization (IDEM) algorithm. Rather than using phase difference measurements that are modeled by a mixture of complex-valued Gaussians, as proposed in the original IDEM framework, we propose to use time difference of arrival measurements in multiple subbands and model them by a mixture of real-valued truncated Gaussians. Moreover, we propose to first filter the measurements in order to reduce the effect of the multipath conditions. The proposed method is evaluated using both simulated data and real-life recordings.

KW - Precedence effect

KW - auditory scene analysis

KW - distributed expectation-maximization

KW - incremental expectation-maximization

KW - multi-path

KW - onset dominance

KW - sound source localization

KW - spectral masking

KW - time difference of arrival

KW - truncated Gaussian

UR - http://www.scopus.com/inward/record.url?scp=85040080751&partnerID=8YFLogxK

U2 - 10.1109/TASLP.2017.2788198

DO - 10.1109/TASLP.2017.2788198

M3 - ???researchoutput.researchoutputtypes.contributiontojournal.article???

AN - SCOPUS:85040080751

SN - 2329-9290

VL - 26

SP - 682

EP - 695

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

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

IS - 3

M1 - 8241767

ER -

Distributed Expectation-Maximization Algorithm for Speaker Localization in Reverberant Environments

Abstract

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this