|
Forensic Phonetics
In phonetics the research concerns the spoken language. This includes several issues like acoustic features of speech, speech production (how sounds are formed in the speech aparatus), perception of speech and recognition by humans and computer programs, acquiring articulatory features and understanding (how one learns articulation and understanding of a language as child or adult), systematic relations and regularities in the spoken language (could apply to different languages,accents, differences, changes and so forth).
Forensic phonetics is the utilisation of and research on language usage in judicial matters. There are many fields of application such as speaker comparison (also referred to as, somewhat misleading, voice identifacation), transcribing of recorded speech or speaker profiling of speakers/voices.
Voxalys AB has since 2005 written forensic phonetic expert estimations concerning criminal cases for police, prosecutors and defence attorneys. Assignments have also been conducted on behalf of "Statens Kriminaltekniska Laboratorium" (SKL) where Voxalys AB has acted as expert consultants in both District and Supreme Court.
Voice comparison and forensic transcription, which are conducted in our forensic speech and acoustic laboratory, contains three analysis parts:
1. Auditory phonetic analysis. Pairwise comparison and editing/transcribing of sound.
2. Acoustic analysis and comparison of vowels, consonants, transitions and prosody.
3. Semi-automatic analysis.
a) Biometric analysis for voice comparison through Mistral (see link below). Comparative population data is used depending on speech and sound quality (training as well as comparison with more than 10.000 speakers is available).
b) Speech recognition (phon- or sound combination recognition through so called 'forced alignment'). Developed through HTK (see link below).
Further information can be found at (in swedish): talaridentifiering.se...
We also offer transcription and translation of a number of different languages. For more information: contact us...
Related links:
Talaridentifiering.se
IPA, utvecklat av International Phonetic Association
IAFPA - International Association for Forensic Phonetics and Acoustics
Mistral - Source code system for biometrics
HTK - The Hidden Markov Model Toolkit
SKL - Statens Kriminaltekniska Laboratorium
Praat - software
Audacity - software
Wavesurfer - software
Methods
What is evaluated in a forensic phonetic report is based on an auditory-phonetic and acoustic-instrumental investigation. Initially an auditory evaluation of the spoken material is made through listening to the material. Those recordings that are analogue are then digitalised with a sampling frequency and bit-rate to ensure best possible quality (for example 44100Hz/16-bit resolution).
Consonant and vowelvariations are then transcribed with the International Phonetic Alphabet (IPA, developed by the International Phonetic Association) to capture detailed features in the speech. Guidelines and directions for a report should always abide by the rules stipulated by IAFPA. The various methods are used in accordance with the particular cases. Regarding voice comparison cases, a biometric voice analysis is also conducted, that is, in the cases the quality of the material allows it. This is done with an automatic system applied to certain chosen parts of the material. In those cases the open source system Mistral id used for biometrics. Mel frequency coefficients (MFCC) and Gaussian Mixture Models (GMM) are used to determine likelihood ratio for the voice belonging to the same speaker vs. the likelihood the speech sample comes from another speaker in a given population (population data that are being compared with comes from accent databases http://www.swedia.nu, forensic material or other telephone recordings depending on the nature of the recorded samples). (Bonastre et al., 2005)
Acoustic analysis (spectrographic and spectral) are then performed to refine specific comparisons of vowels and consonants and to compare with acoustic parameters like fundamental pitch (F0), intonational patterns, average long-time spectrum, formants, MFCC and more. Which acoustic parameters that are being compared depends in large on sound quality regarding format but also on background noise and noise.
Acoustic measurements and analysis are conducted with several software programs like Praat, Audacity and Wavesurfer.
The acoustic analysis follows recommendations from Great Brittain - Northern Ireland Criminal Appeal Court ruling R - v - Anthony O'Doherty (19.iv.02: NICB 3173).
|
|
News
2009-07-10 - voxalys.se will undergo cosmetic operations over the coming weeks!
2009-04-10 - Voxalys 1.3 in Mac-version! All functions including documenting of analysis to a swift web-based analysis template is now also available in Mac-version!
2008-10-02 - The dept. of logopedics, The Sahlgren Academy purchase copy of Voxalys 1.3.
2008-09-01 - Linköping University buys Voxalys Basic.
2008-03-25 - Södra Älvsborg Hospitals logopedes buys Voxalys 1.3.
2007-06-01 - Voxalys 1.3 is used by Christina Lindhe, authorised logopede with solid experience. Since 2003 she is the manager of the business "Röstkonsult Christina Lindhe".
 |
| Subscribe to Voxalys e-mail |
|
