Selective GMM em for telephone diarization

Tal Levy; Itsik Lapidot

doi:10.1109/ICSEE.2016.7806055

Selective GMM em for telephone diarization

Tal Levy
, Itsik Lapidot

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

Abstract

In this paper, we present a novel approach to optimizing the Gaussian-mixture-model (GMM) training phase for a Speaker Diarization System based on hidden distortion models (HDMs). Generally with HDMs, emission models and the transition costs adapts to decrease the overall cost at each iteration. In the current work we use GMM as an emission model. However, in speaker diarization, the real goal is improving the diarization error rate (DER). We used an existing HDM based speaker diarization system and change it by adapting the GMM-EM algorithm to be more selective. Instead of maximizing the likelihood, the new algorithm will emphasize different events by selectively focusing on the K highest posteriors for likelihoods estimation. A proper choice of K leads to a relative improvement of - 5.5%.

Original language	English
Title of host publication	2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016
Publisher	Institute of Electrical and Electronics Engineers
ISBN (Electronic)	9781509021529
DOIs	https://doi.org/10.1109/ICSEE.2016.7806055
State	Published - 4 Jan 2017
Externally published	Yes
Event	2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016 - Eilat, Israel Duration: 16 Nov 2016 → 18 Nov 2016

Publication series

Name	2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016

Conference

Conference	2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016
Country/Territory	Israel
City	Eilat
Period	16/11/16 → 18/11/16

ASJC Scopus subject areas

Computer Science Applications
Hardware and Architecture
Artificial Intelligence
Computer Networks and Communications
Electrical and Electronic Engineering

Access to Document

10.1109/ICSEE.2016.7806055

Cite this

@inproceedings{0fee6b4bbb5d47aa9a30194472a9eb54,

title = "Selective GMM em for telephone diarization",

abstract = "In this paper, we present a novel approach to optimizing the Gaussian-mixture-model (GMM) training phase for a Speaker Diarization System based on hidden distortion models (HDMs). Generally with HDMs, emission models and the transition costs adapts to decrease the overall cost at each iteration. In the current work we use GMM as an emission model. However, in speaker diarization, the real goal is improving the diarization error rate (DER). We used an existing HDM based speaker diarization system and change it by adapting the GMM-EM algorithm to be more selective. Instead of maximizing the likelihood, the new algorithm will emphasize different events by selectively focusing on the K highest posteriors for likelihoods estimation. A proper choice of K leads to a relative improvement of - 5.5\%.",

author = "Tal Levy and Itsik Lapidot",

note = "Publisher Copyright: {\textcopyright} 2016 IEEE.; 2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016 ; Conference date: 16-11-2016 Through 18-11-2016",

year = "2017",

month = jan,

day = "4",

doi = "10.1109/ICSEE.2016.7806055",

language = "English",

series = "2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016",

publisher = "Institute of Electrical and Electronics Engineers",

booktitle = "2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016",

address = "United States",

}

Levy, T & Lapidot, I 2017, Selective GMM em for telephone diarization. in 2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016., 7806055, 2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016, Institute of Electrical and Electronics Engineers, 2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016, Eilat, Israel, 16/11/16. https://doi.org/10.1109/ICSEE.2016.7806055

Selective GMM em for telephone diarization. / Levy, Tal; Lapidot, Itsik.
2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016. Institute of Electrical and Electronics Engineers, 2017. 7806055 (2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016).

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

TY - GEN

T1 - Selective GMM em for telephone diarization

AU - Levy, Tal

AU - Lapidot, Itsik

PY - 2017/1/4

Y1 - 2017/1/4

N2 - In this paper, we present a novel approach to optimizing the Gaussian-mixture-model (GMM) training phase for a Speaker Diarization System based on hidden distortion models (HDMs). Generally with HDMs, emission models and the transition costs adapts to decrease the overall cost at each iteration. In the current work we use GMM as an emission model. However, in speaker diarization, the real goal is improving the diarization error rate (DER). We used an existing HDM based speaker diarization system and change it by adapting the GMM-EM algorithm to be more selective. Instead of maximizing the likelihood, the new algorithm will emphasize different events by selectively focusing on the K highest posteriors for likelihoods estimation. A proper choice of K leads to a relative improvement of - 5.5%.

AB - In this paper, we present a novel approach to optimizing the Gaussian-mixture-model (GMM) training phase for a Speaker Diarization System based on hidden distortion models (HDMs). Generally with HDMs, emission models and the transition costs adapts to decrease the overall cost at each iteration. In the current work we use GMM as an emission model. However, in speaker diarization, the real goal is improving the diarization error rate (DER). We used an existing HDM based speaker diarization system and change it by adapting the GMM-EM algorithm to be more selective. Instead of maximizing the likelihood, the new algorithm will emphasize different events by selectively focusing on the K highest posteriors for likelihoods estimation. A proper choice of K leads to a relative improvement of - 5.5%.

UR - https://www.scopus.com/pages/publications/85014161630

U2 - 10.1109/ICSEE.2016.7806055

DO - 10.1109/ICSEE.2016.7806055

M3 - Conference contribution

AN - SCOPUS:85014161630

T3 - 2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016

BT - 2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016

PB - Institute of Electrical and Electronics Engineers

T2 - 2016 IEEE International Conference on the Science of Electrical Engineering, ICSEE 2016

Y2 - 16 November 2016 through 18 November 2016

ER -

Selective GMM em for telephone diarization

Abstract

Publication series

Conference

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this