LR-based forensic comparison under severe test-data scarcity

Yuko Kinoshita; Michael Wagner

LR-based forensic comparison under severe test-data scarcity

Yuko Kinoshita, Michael Wagner

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

Abstract

This study sets out to find the most reliable method for loglikelihood-ratio (LLR) calculation under severe data scarcity, which is typical of forensic voice comparison casework. We compared the performances of three types of speaker modelling, namely a single Gaussian model, Gaussian Mixture Models (GMM) of different complexity, and a Multivariate Kernel Density Model (MVKD), using two and threedimensional formant frequency feature vectors extracted from /iː/ vowels. We varied the number of tokens used in the offender dataset from 2 to 6. We find that calibration of the systems was critical for dependable evaluation with all the systems tested and that the MVKD model outperformed Gaussian models in most cases.

Original language	English
Title of host publication	Interpeech 2014
Place of Publication	Singapore
Publisher	International Speech Communication Association
Pages	16-19
Edition	Peer Reviewed
Publication status	Published - 2014
Event	Annual Conference of the International Speech Communication Association INTERSPEECH 2014 - Singapore, Singapore Duration: 1 Jan 2014 → … http://www.isca-speech.org/archive/interspeech_2014

Conference

Conference	Annual Conference of the International Speech Communication Association INTERSPEECH 2014
Country/Territory	Singapore
Period	1/01/14 → …
Other	September 14-18 2014
Internet address	http://www.isca-speech.org/archive/interspeech_2014

Cite this

@inproceedings{16bb124cd93843e7b659bb3b50cae347,

title = "LR-based forensic comparison under severe test-data scarcity",

abstract = "This study sets out to find the most reliable method for loglikelihood-ratio (LLR) calculation under severe data scarcity, which is typical of forensic voice comparison casework. We compared the performances of three types of speaker modelling, namely a single Gaussian model, Gaussian Mixture Models (GMM) of different complexity, and a Multivariate Kernel Density Model (MVKD), using two and threedimensional formant frequency feature vectors extracted from /iː/ vowels. We varied the number of tokens used in the offender dataset from 2 to 6. We find that calibration of the systems was critical for dependable evaluation with all the systems tested and that the MVKD model outperformed Gaussian models in most cases.",

author = "Yuko Kinoshita and Michael Wagner",

year = "2014",

language = "English",

pages = "16--19",

booktitle = "Interpeech 2014",

publisher = "International Speech Communication Association",

edition = "Peer Reviewed",

note = "Annual Conference of the International Speech Communication Association INTERSPEECH 2014 ; Conference date: 01-01-2014",

url = "http://www.isca-speech.org/archive/interspeech\_2014",

}

TY - GEN

T1 - LR-based forensic comparison under severe test-data scarcity

AU - Kinoshita, Yuko

AU - Wagner, Michael

PY - 2014

Y1 - 2014

N2 - This study sets out to find the most reliable method for loglikelihood-ratio (LLR) calculation under severe data scarcity, which is typical of forensic voice comparison casework. We compared the performances of three types of speaker modelling, namely a single Gaussian model, Gaussian Mixture Models (GMM) of different complexity, and a Multivariate Kernel Density Model (MVKD), using two and threedimensional formant frequency feature vectors extracted from /iː/ vowels. We varied the number of tokens used in the offender dataset from 2 to 6. We find that calibration of the systems was critical for dependable evaluation with all the systems tested and that the MVKD model outperformed Gaussian models in most cases.

AB - This study sets out to find the most reliable method for loglikelihood-ratio (LLR) calculation under severe data scarcity, which is typical of forensic voice comparison casework. We compared the performances of three types of speaker modelling, namely a single Gaussian model, Gaussian Mixture Models (GMM) of different complexity, and a Multivariate Kernel Density Model (MVKD), using two and threedimensional formant frequency feature vectors extracted from /iː/ vowels. We varied the number of tokens used in the offender dataset from 2 to 6. We find that calibration of the systems was critical for dependable evaluation with all the systems tested and that the MVKD model outperformed Gaussian models in most cases.

M3 - Conference Paper

SP - 16

EP - 19

BT - Interpeech 2014

PB - International Speech Communication Association

CY - Singapore

T2 - Annual Conference of the International Speech Communication Association INTERSPEECH 2014

Y2 - 1 January 2014

ER -

LR-based forensic comparison under severe test-data scarcity

Abstract

Conference

Fingerprint

Cite this