Abstracts Submitted: 510
Number of Users: 710
View Abstracts Submitted
Back to home Page
Forensic speaker identification is the process of comparison of unknown speech samples with the known ones for the purpose of legal requirement. Various studies are being carried out on the acoustic parameters extracted from the voice signals to identify a speaker and on that basis the forensic speech scientists are making their opinions. Most of the studies were carried out in very controlled manner with limited samples to avoid the effects due to the variations. It is obvious that it is very tough to make definite opinion in forensic speaker identifications as the dynamic variations of acoustic features are inherent nature of voice. In forensic scenario the process of speaker identification examination becomes more challenging, because the voice acoustic parameters are not only varying for different speakers but also varies even for the same speaker under different physical, psychological conditions. Here, the acoustic features of recorded voice signals of eleven Kongu Tamil language speakers in two modes namely normal reading and natural speaking modes were studied using Computerised Speech Lab(CSL). The sample words are taken in three different ways on the basis of vowel environment and mode of recording. Wide band spectrum was generated using CSL and Formant frequencies were measured for the six prominent vowels considered for this study. Scaling was worked out on the basis of matching of points in the generated pattern using radar diagram. The scaling was also analysed to find out its application in speaker identification. The results are combination of objective and subjective opinions, generally adopted in forensic speaker identification. The present study shows that to get the similar pattern at least 66.6% of values should be matching. According to vowel environment, variations observed in scaling from positive to probable matching. The ratio of 'F2' to 'F1' was found more resolving characteristics than average value of 'F2'. The advantage of this method is that most of the subjective part has been converted in objective one and the percentage of failure observed is only 0.03%. Hence, this will be simple and first hand method to make the opinion for the purpose of speaker identification. The readers may create their own system of vowels patterns according to their language of interest.
© Copyright 2017 All Rights Reserved