Audio forensics: a voice identification investigation

Abstract

The development of computer technology has result in demand for more effective intelligent computer program. One of the areas is speaker identification (SI). SI is the process of identifying the speaker based on the characteristics contained in their speech waves. This process can be used in forensic investigation to recognize voice of suspected criminal. Nowadays, a lot of methods can be used to perform speaker identification. Nevertheless, the accuracy of these methods is different according to its algorithm that being used as well as the analyzing of the data. One of methods that can be used for speaker recognition is wavelet transform (WT). WT divided into two methods; discrete wavelet transform (DWT) method and continuous wavelet transform method. This research focused in the implementation, development and analyzing the accuracy of DWT in identifying voice. The experiment is conducted to recognize the spoken person and this is done in four different approaches: recognition based on a single predefined spoken word with normal voice, recognition based on a single predefined spoken word with non-normal voice (with nose closed), recognition based a on multiple spoken words including the predefined word and recognition based on single predefined word but with different tone frequency. The results obtained are voice with changed frequency such as in experiment two and three gives accuracy below 50 percent and for voice with normal frequency like in experiment one and four gives the accuracy above 80 percent. However DWT gives satisfactory result if the voice frequency is normal.