Rouf, Roudhotul Jannah (2020) Identifikasi Forensik Suara Menggunakan Metode Joint Factor Analysis Dan I-Vector. Other thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
02311640000011-Undergraduate_Thesis.pdf Download (4MB) | Preview |
Abstract
Forensik suara ucap adalah proses untuk menentukan kecocokan identitas antara suara seseorang (known speaker) dengan suara yang akan diselidiki (suspect speaker). Untuk meningkatkan keakuratan dalam analisa forensik suara digunakan gabungan 2 metode forensik. Penelitian ini bertujuan untuk membuat sistem identifikasi forensik suara ucap dengan metode Joint Factor Analysis dan i-vector. Pendekatan forensik dilakukan dengan menambahkan sinyal noise dengan nilai SNR tertentu sebagai representasi situasi penyadapan untuk mengukur performansi identifikasi suara. Verifikasi dilakukan dengan 2 pengujian yaitu perbandingan speaker yang sama dan verifikasi antara speaker yang berbeda. Klasifikasi pada verifikasi penutur menggunakan i-vector dilakukan dengan membandingkan model i-vector tes dan target. Kedua model tersebut dihitung kemiripan vectornya menggunakan cosine similarity score. Proses pengujian pada performa program ditunjukkan oleh nilai error equal rate. Sedangkan sensivitas sistem ditunjukkan oleh nilai threshold. Hasil yang diperoleh menunjukkan bahwa nilai EER dataset Graz (1,8%) lebih rendah dibandingkan dengan dataset Indonesia (10%).
====================================================================================================================
Voice forensics is the process of determining the identity match between a person's voice (known speaker) with the voice to be investigated (suspect speaker). To improve accuracy in sound forensic analysis a combination of 2 forensic methods is used. This study aims to create a sound forensic identification system using the Joint Factor Analysis and i-vector methods. The forensic approach is carried out by adding noise signals with certain SNR values as a representation of the tapping situation to measure the performance of sound identification. Verification is done by 2 tests namely the comparison of the same speaker and verification between different speakers. Classification of speaker verification using i-vector is done by comparing the i-vector test model and the target. Both of the models in the vector similarity were calculated using a cosine similarity score. The testing process on program performance is indicated by the error equal rate value. While the sensitivity of the system is shown by the value of the threshold. The results obtained indicate that the Graz dataset EER value (1,8%) is lower than the Indonesian dataset (10%).
| Item Type: | Thesis (Other) |
|---|---|
| Uncontrolled Keywords: | Forensik, Joint Factor Analysis, i-vector, Cosine Similarity Score, Error Equal Rate. Forensic, Joint Factor Analysis, i-vector, Cosine Similarity Score, Error Equal Rate |
| Subjects: | Q Science > QA Mathematics > QA76.9.A25 Computer security. Digital forensic. Data encryption (Computer science) |
| Divisions: | Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis |
| Depositing User: | Roudhotul Jannah Rouf |
| Date Deposited: | 27 Aug 2020 08:52 |
| Last Modified: | 25 Nov 2023 15:35 |
| URI: | http://repository.its.ac.id/id/eprint/81295 |
Actions (login required)
 |
View Item |