Null-Hypothesis LLR: A proposal for Forensic Automatic Speaker Recognition

Yosef A. Solewicz; Michael Jessen; David van der Vloed

doi:10.21437/Interspeech.2017-1023

Null-Hypothesis LLR: A proposal for Forensic Automatic Speaker Recognition

Yosef A. Solewicz
, Michael Jessen
, David van der Vloed

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

7 Citations (Scopus)

Abstract

A new method named Null-Hypothesis LLR (H₀LLR) is proposed for forensic automatic speaker recognition. The method takes into account the fact that forensically realistic data are difficult to collect and that inter-individual variation is generally better represented than intra-individual variation.
According to the proposal, intra-individual variation is modeled as a projection from case-customized inter-individual variation. Calibrated log Likelihood Ratios (LLR) that are calculated on the basis of the H₀LLR method were tested on two corpora of forensically-founded telephone interception test sets, German-based GFS 2.0 and Dutch-based NFIFRITS. Five automatic speaker recognition systems were tested based on the scores or the LLRs provided by these systems which form the input to H₀LLR. Speaker discrimination and calibration performance of H₀LLR is comparable to the performance indices of the system-internal LLR calculation methods. This shows that external data and strategies that work with data outside the forensic domain and without case customization are not necessary. It is also shown that H₀LLR leads to a reduction in the diversity of LLR output patterns of different automatic systems. This is important for the credibility of the Likelihood Ratio framework in forensics, and its application in forensic automatic speaker recognition in particular.

Original language	English
Title of host publication	Proceedings of the Annual Conference of the International Speech Communication Association, Interspeech 2017
Editors	F. Lacerda, S. Strombergsson, M. Wlodarczak, M. Heldner, J. Gustafson, D. House
Publisher	ISCA
Pages	2849-2853
Number of pages	5
DOIs	https://doi.org/10.21437/Interspeech.2017-1023
Publication status	Published - 24 Aug 2017
Externally published	Yes
Event	18th Annual Conference of the International Speech Communication Association, Interspeech 2017: Situated interaction - Stockholm, Sweden Duration: 20 Aug 2017 → 24 Aug 2017 Conference number: 18 https://researchr.org/publication/interspeech-2017

Publication series

Name	Proceedings of the Annual Conference of the International Speech Communication Association
Publisher	ISCA
Volume	2017-August
ISSN (Print)	2308-457X

Conference

Conference	18th Annual Conference of the International Speech Communication Association, Interspeech 2017
Abbreviated title	Interspeech 2017
Country/Territory	Sweden
City	Stockholm
Period	20/08/17 → 24/08/17
Internet address	https://researchr.org/publication/interspeech-2017

Keywords

n/a OA procedure
forensic speech science
likelihood ratios
calibration
automatic speaker recognition

Access to Document

10.21437/Interspeech.2017-1023

Cite this

Solewicz, Y. A., Jessen, M., & van der Vloed, D. (2017). Null-Hypothesis LLR: A proposal for Forensic Automatic Speaker Recognition. In F. Lacerda, S. Strombergsson, M. Wlodarczak, M. Heldner, J. Gustafson, & D. House (Eds.), Proceedings of the Annual Conference of the International Speech Communication Association, Interspeech 2017 (pp. 2849-2853). (Proceedings of the Annual Conference of the International Speech Communication Association; Vol. 2017-August). ISCA. https://doi.org/10.21437/Interspeech.2017-1023

Solewicz, Yosef A. ; Jessen, Michael ; van der Vloed, David. / Null-Hypothesis LLR : A proposal for Forensic Automatic Speaker Recognition. Proceedings of the Annual Conference of the International Speech Communication Association, Interspeech 2017. editor / F. Lacerda ; S. Strombergsson ; M. Wlodarczak ; M. Heldner ; J. Gustafson ; D. House. ISCA, 2017. pp. 2849-2853 (Proceedings of the Annual Conference of the International Speech Communication Association).

@inproceedings{4117a4cd93d445f3b7860624094af48d,

title = "Null-Hypothesis LLR: A proposal for Forensic Automatic Speaker Recognition",

abstract = "A new method named Null-Hypothesis LLR (H0LLR) is proposed for forensic automatic speaker recognition. The method takes into account the fact that forensically realistic data are difficult to collect and that inter-individual variation is generally better represented than intra-individual variation. According to the proposal, intra-individual variation is modeled as a projection from case-customized inter-individual variation. Calibrated log Likelihood Ratios (LLR) that are calculated on the basis of the H0LLR method were tested on two corpora of forensically-founded telephone interception test sets, German-based GFS 2.0 and Dutch-based NFIFRITS. Five automatic speaker recognition systems were tested based on the scores or the LLRs provided by these systems which form the input to H0LLR. Speaker discrimination and calibration performance of H0LLR is comparable to the performance indices of the system-internal LLR calculation methods. This shows that external data and strategies that work with data outside the forensic domain and without case customization are not necessary. It is also shown that H0LLR leads to a reduction in the diversity of LLR output patterns of different automatic systems. This is important for the credibility of the Likelihood Ratio framework in forensics, and its application in forensic automatic speaker recognition in particular.",

keywords = "n/a OA procedure, forensic speech science, likelihood ratios, calibration, automatic speaker recognition",

author = "Solewicz, \{Yosef A.\} and Michael Jessen and \{van der Vloed\}, David",

note = "Copyright {\textcopyright} 2017 ISCA. ; 18th Annual Conference of the International Speech Communication Association, Interspeech 2017 : Situated interaction, Interspeech 2017 ; Conference date: 20-08-2017 Through 24-08-2017",

year = "2017",

month = aug,

day = "24",

doi = "10.21437/Interspeech.2017-1023",

language = "English",

series = "Proceedings of the Annual Conference of the International Speech Communication Association",

publisher = "ISCA",

pages = "2849--2853",

editor = "F. Lacerda and S. Strombergsson and M. Wlodarczak and M. Heldner and J. Gustafson and House, \{D. \}",

booktitle = "Proceedings of the Annual Conference of the International Speech Communication Association, Interspeech 2017",

url = "https://researchr.org/publication/interspeech-2017",

}

Solewicz, YA, Jessen, M & van der Vloed, D 2017, Null-Hypothesis LLR: A proposal for Forensic Automatic Speaker Recognition. in F Lacerda, S Strombergsson, M Wlodarczak, M Heldner, J Gustafson & D House (eds), Proceedings of the Annual Conference of the International Speech Communication Association, Interspeech 2017. Proceedings of the Annual Conference of the International Speech Communication Association, vol. 2017-August, ISCA, pp. 2849-2853, 18th Annual Conference of the International Speech Communication Association, Interspeech 2017, Stockholm, Sweden, 20/08/17. https://doi.org/10.21437/Interspeech.2017-1023

Null-Hypothesis LLR: A proposal for Forensic Automatic Speaker Recognition. / Solewicz, Yosef A.; Jessen, Michael; van der Vloed, David.
Proceedings of the Annual Conference of the International Speech Communication Association, Interspeech 2017. ed. / F. Lacerda; S. Strombergsson; M. Wlodarczak; M. Heldner; J. Gustafson; D. House. ISCA, 2017. p. 2849-2853 (Proceedings of the Annual Conference of the International Speech Communication Association; Vol. 2017-August).

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

TY - GEN

T1 - Null-Hypothesis LLR

T2 - 18th Annual Conference of the International Speech Communication Association, Interspeech 2017

AU - Solewicz, Yosef A.

AU - Jessen, Michael

AU - van der Vloed, David

N1 - Conference code: 18

PY - 2017/8/24

Y1 - 2017/8/24

N2 - A new method named Null-Hypothesis LLR (H0LLR) is proposed for forensic automatic speaker recognition. The method takes into account the fact that forensically realistic data are difficult to collect and that inter-individual variation is generally better represented than intra-individual variation. According to the proposal, intra-individual variation is modeled as a projection from case-customized inter-individual variation. Calibrated log Likelihood Ratios (LLR) that are calculated on the basis of the H0LLR method were tested on two corpora of forensically-founded telephone interception test sets, German-based GFS 2.0 and Dutch-based NFIFRITS. Five automatic speaker recognition systems were tested based on the scores or the LLRs provided by these systems which form the input to H0LLR. Speaker discrimination and calibration performance of H0LLR is comparable to the performance indices of the system-internal LLR calculation methods. This shows that external data and strategies that work with data outside the forensic domain and without case customization are not necessary. It is also shown that H0LLR leads to a reduction in the diversity of LLR output patterns of different automatic systems. This is important for the credibility of the Likelihood Ratio framework in forensics, and its application in forensic automatic speaker recognition in particular.

AB - A new method named Null-Hypothesis LLR (H0LLR) is proposed for forensic automatic speaker recognition. The method takes into account the fact that forensically realistic data are difficult to collect and that inter-individual variation is generally better represented than intra-individual variation. According to the proposal, intra-individual variation is modeled as a projection from case-customized inter-individual variation. Calibrated log Likelihood Ratios (LLR) that are calculated on the basis of the H0LLR method were tested on two corpora of forensically-founded telephone interception test sets, German-based GFS 2.0 and Dutch-based NFIFRITS. Five automatic speaker recognition systems were tested based on the scores or the LLRs provided by these systems which form the input to H0LLR. Speaker discrimination and calibration performance of H0LLR is comparable to the performance indices of the system-internal LLR calculation methods. This shows that external data and strategies that work with data outside the forensic domain and without case customization are not necessary. It is also shown that H0LLR leads to a reduction in the diversity of LLR output patterns of different automatic systems. This is important for the credibility of the Likelihood Ratio framework in forensics, and its application in forensic automatic speaker recognition in particular.

KW - n/a OA procedure

KW - forensic speech science

KW - likelihood ratios

KW - calibration

KW - automatic speaker recognition

UR - https://www.scopus.com/pages/publications/85039170679

U2 - 10.21437/Interspeech.2017-1023

DO - 10.21437/Interspeech.2017-1023

M3 - Conference contribution

T3 - Proceedings of the Annual Conference of the International Speech Communication Association

SP - 2849

EP - 2853

BT - Proceedings of the Annual Conference of the International Speech Communication Association, Interspeech 2017

A2 - Lacerda, F.

A2 - Strombergsson, S.

A2 - Wlodarczak, M.

A2 - Heldner, M.

A2 - Gustafson, J.

A2 - House, D.

PB - ISCA

Y2 - 20 August 2017 through 24 August 2017

ER -

Solewicz YA, Jessen M, van der Vloed D. Null-Hypothesis LLR: A proposal for Forensic Automatic Speaker Recognition. In Lacerda F, Strombergsson S, Wlodarczak M, Heldner M, Gustafson J, House D, editors, Proceedings of the Annual Conference of the International Speech Communication Association, Interspeech 2017. ISCA. 2017. p. 2849-2853. (Proceedings of the Annual Conference of the International Speech Communication Association). doi: 10.21437/Interspeech.2017-1023

Null-Hypothesis LLR: A proposal for Forensic Automatic Speaker Recognition

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this