Research Committee | Previous Projects
The Long Term Cumulative Formant Distribution in Forensic Voice Analysis
Catalin Grigoras, National Institute of Forensic Expertise, Bucharest, Romania
Catalina Software Now Available
The Catalina Forensic Audio Toolbox software which has been developed with the help of an IAFPA research grant is available on the download page of Catalin Grigoras' web site.
Project Summary
This study is an extension of the research on the Long Term Cumulative Formant Distribution (LTCFD) in forensic voice analysis. The aim is to determine both the advantages and limits of this kind of analysis in forensic cases. Two Romanian language speaker databases will be used: lab quality recordings and speech from real cases. For the material from real cases, only good quality audio samples (lab controlled conditions, digital recordings) will be analyzed.
For each recording an acoustic profile (consisting of F0, LTAS, F1-2-3), LTCFD and formant space will be measured and derived. LTF (as is described in Hollien, H., 2000, Forensic Voice Identification, Academic Press) and LTCFD contain speaker particularities through which it is possible to quantify auditorily detectable similarities and differences. A likelihood ratio (LR as is presented in Rose, P., 2002, Forensic Speaker Identification, Taylor & Francis) approach will also be considered and the results will be discussed.
The application of LTCFD to both within and between speaker variability will be analyzed too. The correlation of results will be presented and used to describe the benefits and limits of this kind of analysis.
DCLive Forensics will be used to acquire the speech signals and/or to reduce noise without modifying formants. Wavesurfer will be the software used to extract F0, LTAS, F1-2-3 values and Catalina Forensic Audio Toolbox will be used to analyze them.
A version of Catalina Forensic Audio Software will be made available to IAFPA members interested in this topic. Members will be able to make their own tests to further understand LTCFD. For this version the database is limited to 5 speakers. Then, the possibility to use Catalina Toolbox in a complex methodology for forensic voice analysis can be explored.
Further directions to upgrade Catalina Toolbox (e.g. increase number of speakers in database) will follow this study and this software version (e.g. the addition of Long Term Average Spectra Histogram).