Manipulation of human voice: Forensic investigation with acoustic analyses

Speech recordings that are subject to criminal investigations serve as audible witnesses to the crime. These witnesses are just as important as physical evidence obtained from the crime scene due to their ability to be easily concealed and transmitted. The recorded speech sample/s provide quantitative data for solving the crime and thus help establish the innocence or guilt of the individual in criminal cases. The aim of this study is to determine how formant frequencies of the vowels that are effective when the speech signal manipulates the individual's voice are affected. Additionally, the study aims to reveal the similarities and differences between the suspect's manipulated voice recordings and the original voice recording in terms of formant frequencies. This is to guide voice experts in identifying the identity of unknown voice recordings for the prosecution and the courts. In this study, a total of 50 healthy volunteers, 25 females and 25 males between the ages of 20-55, participated. Five different voice recordings were taken for each individual. These five different recordings consist of the original voice recording without any intervention and four manipulated recordings where one of the voice concealment methods (nasal obstruction, mask, handkerchief, and whispering) was used each time. Two different tasks were prepared for the voice recordings: 1) lists of prepared words and 2) reading a paragraph. Audacity software was used for the voice recording, and custom-designed Praat software was used for voice analysis. The F0, F1, F2, F3 formant frequencies of eight vowels (/a/, /e/, /ı/, /i/, /o/, /ö/, /u/, and /ü/) were measured, and the data obtained from the same point were compared. Furthermore, spectrographic analysis of the sentence "He spends the whole day with his teacher and friends" was performed. The obtained spectrograms were examined in terms of formant frequency distribution, noise components, and periodic status of frequencies. Statistical analyses were performed using the IBM SPSS for Windows Version 23.0 package program. As a result, for the identification of identity in a single-syllable (three-letter) sentence set concerning "formant frequency values," reliable formant frequencies "F0" (for reliable phonemes /a/, /e/, /ı/, /o/, /ö/, /u/, and /ü/) and "F1" (for reliable phonemes /a/, /e/, /i/, /o/, /ö/, /u/, and /ü/) were found. For the identification of identity in speech-related formant frequency values, "F0" (for reliable phonemes /a/, /e/, /ı/, /o/, /ö/, and /ü/) and "F1" (for reliable phonemes /a/, /e/, /ı/, /o/, /ö/, and /ü/) formant frequencies are used. Spectrogram typifications provide a documented idea of how much intervention has been made in the natural characteristics of the voice to forensic audio experts.