A new penalty term for the BIC with respect to speaker diarization

In this paper we examine a new penalty term for the Bayesian Information Criterion (BIC) that is suited to the problem of speaker diarization. Based on our previous approach of penalizing each cluster only with its effective sample size - an approach we called segmental - we propose a stricter penal...

Full description

Bibliographic Details
Main Authors: Stafylakis, Themos, Tzimiropoulos, Georgios, Katsouros, Vassilis, Carayannis, George
Format: Conference or Workshop Item
Published: 2010
Subjects:
Online Access:https://eprints.nottingham.ac.uk/31422/
_version_ 1848794197730000896
author Stafylakis, Themos
Tzimiropoulos, Georgios
Katsouros, Vassilis
Carayannis, George
author_facet Stafylakis, Themos
Tzimiropoulos, Georgios
Katsouros, Vassilis
Carayannis, George
author_sort Stafylakis, Themos
building Nottingham Research Data Repository
collection Online Access
description In this paper we examine a new penalty term for the Bayesian Information Criterion (BIC) that is suited to the problem of speaker diarization. Based on our previous approach of penalizing each cluster only with its effective sample size - an approach we called segmental - we propose a stricter penalty term. The criterion we derive retains the main property of the Segmental-BIC, i.e. it approximates the evidence of overall partitions of the data and simultaneously leads to a pairwise dissimilarity measure that is completely defined by the pair of clusters in question. The experimental results show significant improvement in diarization accuracy on the ESTER benchmark.
first_indexed 2025-11-14T19:12:22Z
format Conference or Workshop Item
id nottingham-31422
institution University of Nottingham Malaysia Campus
institution_category Local University
last_indexed 2025-11-14T19:12:22Z
publishDate 2010
recordtype eprints
repository_type Digital Repository
spelling nottingham-314222020-05-04T20:25:56Z https://eprints.nottingham.ac.uk/31422/ A new penalty term for the BIC with respect to speaker diarization Stafylakis, Themos Tzimiropoulos, Georgios Katsouros, Vassilis Carayannis, George In this paper we examine a new penalty term for the Bayesian Information Criterion (BIC) that is suited to the problem of speaker diarization. Based on our previous approach of penalizing each cluster only with its effective sample size - an approach we called segmental - we propose a stricter penalty term. The criterion we derive retains the main property of the Segmental-BIC, i.e. it approximates the evidence of overall partitions of the data and simultaneously leads to a pairwise dissimilarity measure that is completely defined by the pair of clusters in question. The experimental results show significant improvement in diarization accuracy on the ESTER benchmark. 2010 Conference or Workshop Item PeerReviewed Stafylakis, Themos, Tzimiropoulos, Georgios, Katsouros, Vassilis and Carayannis, George (2010) A new penalty term for the BIC with respect to speaker diarization. In: ICASSP 2010 - 2010 IEEE International Conference on Acoustics Speech and Signal Processing, 14-19 March 2010, Dallas, USA. Bayes Methods Speaker Recognition Bayesian Information Criterion Cluster Analysis Speaker Diarization http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=5495076
spellingShingle Bayes Methods
Speaker Recognition
Bayesian Information Criterion
Cluster Analysis
Speaker Diarization
Stafylakis, Themos
Tzimiropoulos, Georgios
Katsouros, Vassilis
Carayannis, George
A new penalty term for the BIC with respect to speaker diarization
title A new penalty term for the BIC with respect to speaker diarization
title_full A new penalty term for the BIC with respect to speaker diarization
title_fullStr A new penalty term for the BIC with respect to speaker diarization
title_full_unstemmed A new penalty term for the BIC with respect to speaker diarization
title_short A new penalty term for the BIC with respect to speaker diarization
title_sort new penalty term for the bic with respect to speaker diarization
topic Bayes Methods
Speaker Recognition
Bayesian Information Criterion
Cluster Analysis
Speaker Diarization
url https://eprints.nottingham.ac.uk/31422/
https://eprints.nottingham.ac.uk/31422/