Abstract
Abstract Purpose: There are a number of challenges in forensic speaker identification (FSI) due to the variations in recorded speech samples. Two major variations which uniformly spread in entire speech samples are as follows: variation due to (1) speaking style and (2) vowel environment. The present study is carried out in Tamil language, and here, we study variations in formant frequency due to different speaking styles, i.e., reading and natural spontaneous speech mode. Materials and Methods: Speech samples were recorded in both reading and natural speech mode from 11 persons known to the authors. The present study is based on auditory and spectrographic analysis, a conventional method followed in FSI. The spectrographic analysis and measurement of formant frequencies were carried out by using computerized speech lab. Results: The study was carried out on the basis of Euclidian distances and statistical analysis on formant frequencies, and it was observed that the formant frequencies do not vary significantly during reading mode; however, in natural speech mode, it vary significantly. A detailed study on the effect of change in vowel environment was carried out, and it was revealed that there is a significant variation in formant frequencies due to difference in realization. Conclusion: The results show that the vocal tract geometry of a speaker shall vary significantly due to the difference in speaking style and vowel environment. The results obtained in this study will be useful in modeling text-dependent and text-independent automatic speaker identification systems. Further, there is a scope for similar study in different languages and applications.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call