Publication Details

ROXSD: The ROXANNE Multimodal and Simulated Dataset for Advancing Criminal Investigations

MOTLÍČEK Petr, DIKICI Erinç, MADIKERI Srikanth, RANGAPPA Pradeep, BACKFRIED Gerhard, ROHDIN Johan A., SCHWARZ Petr, KOVÁČ Marek, MALÝ Květoslav, BOBOŠ Dominik, KLAKOW Dietrich and SERGIDOU Eleni Konstantina et al. ROXSD: The ROXANNE Multimodal and Simulated Dataset for Advancing Criminal Investigations. In: Proceedings of Odyssey 2024: The Speaker and Language Recognition Workshop. Québec City: International Speech Communication Association, 2024, pp. 17-24. Available from: https://www.isca-archive.org/odyssey_2024/motlicek24_odyssey.pdf
Czech title
ROXSD: Multimodální simulovaný dataset projektu ROXANNE pro pokrok ve vyšetřování trestných činů
Type
conference paper
Language
english
Authors
Motlíček Petr, doc. Ing., Ph.D. (DCGM FIT BUT)
Dikici Erinç (HENSOLDT)
Madikeri Srikanth (IDIAP)
Rangappa Pradeep (IDIAP)
Backfried Gerhard (SAIL LABS Technology AG)
Rohdin Johan A., Dr. (DCGM FIT BUT)
Schwarz Petr, Ing., Ph.D. (DCGM FIT BUT)
Kováč Marek (Phonexia)
Malý Květoslav, Ing. (Phonexia)
Boboš Dominik (Phonexia)
Klakow Dietrich (UDS)
and others
URL
Keywords

Multimodal and Simulated Dataset, Advancing Criminal Investigations

Abstract

The ROXANNE project, conducted under the European Union's Horizon 2020 Programme, aimed to revolutionize criminal investigations by integrating speech, language, and video technologies with criminal network analysis. Despite the success in technology development, the project faced evaluation challenges due to the scarcity and legal restrictions surround- ing real-world criminal activity datasets. In response, we intro- duce ROXSD, a simulated dataset of communication in orga- nized crime. ROXSD is a set of wiretapped conversations (col- lected through communication service providers) between drug dealing suspects, following a realistic screenplay (incl. realis- tic conditions and constraints of a real investigation) prepared by Law Enforcement Agencies (LEAs). With a focus on multi- modality and multilinguality, the dataset comprises 20 hours of telephone and video conversations involving 104 speakers, and is further aligned with ground-truth annotations for each modal- ity involved, enabling precise evaluation and development of technologies. In addition, the multimodal data are enhanced with metadata and prior knowledge (e.g., suspects' biometric profiles) which is typically available as a result of lawfully in- tercepted communication. This paper introduces ROXSD as a pivotal resource for advancing technology in criminal research (specifically in domain of speech, text and network analysis). ROXSD not only facilitates in the domain of technology devel- opment and evaluation but also showcases the potential of sim- ulated datasets in advancing the field of organized crime analyt- ics, emphasizing the importance of such datasets in the absence of comprehensive real-world alternatives.

Published
2024
Pages
17-24
Proceedings
Proceedings of Odyssey 2024: The Speaker and Language Recognition Workshop
Conference
Odyssey 2024: The Speaker and Language Recognition Workshop, Quebec, Canada, CA
Publisher
International Speech Communication Association
Place
Québec City, CA
DOI
10.21437/odyssey.2024-3
BibTeX 
@INPROCEEDINGS{FITPUB13307,
   author = "Petr Motl\'{i}\v{c}ek and Erin\c{c} Dikici and Srikanth Madikeri and Pradeep Rangappa and Gerhard Backfried and A. Johan Rohdin and Petr Schwarz and Marek Kov\'{a}\v{c} and Kv\v{e}toslav Mal\'{y} and Dominik Bobo\v{s} and Dietrich Klakow and Konstantina Eleni Sergidou and et al.",
   title = "ROXSD: The ROXANNE Multimodal and Simulated Dataset for Advancing Criminal Investigations",
   pages = "17--24",
   booktitle = "Proceedings of Odyssey 2024: The Speaker and Language Recognition Workshop",
   year = 2024,
   location = "Qu\'{e}bec City, CA",
   publisher = "International Speech Communication Association",
   doi = "10.21437/odyssey.2024-3",
   language = "english",
   url = "https://www.fit.vut.cz/research/publication/13307"
}
Back to top