kumoh national institute of technology
Networked Systems Lab.

Kevin Putra Dirgantoro, Gaspard Gashema, Dong-Seong Kim, "A Comparative Study of Cepstral Analysis and MFCC for Speech Steganalysis", 2018 KICS Fall Conference, pp. 36-37, November 17, 2018, Korea University, South Korea (N2 & N8)
By : 관리자
Date : 2018-10-20
Views : 453

This paper performs a comparative study of speech signal with two different methods. The first method is cepstral analysis that performs a low-time liftering in quefrency domain to obtain the characteristics of the vocal tract. The second method is the mel-frequency cepstral coefficient (MFCC) which is used as a good vector to represent the speech signal with 20 mel-filterbank. The objective of this study is to analyze which method has a better accuracy to identify the speech signal with a secret message inside. According to the simulation results, MFCC outperforms cepstral analysis. The value is obtained from statistical features using support vector machine (SVM) for its classification.


Presentation's Picture from KICS Fall Conference 2018




P: For the accuracy still not reach 100% so it means there are still some errors?

K: Yes, so for the accuracy it self can be obtained from calculating the number of files that declared as correct with total files that being used. In this research not all of the files that declared, I think it is because of the machine learning that we used.


P: So, you're still using a simple and old machine learning. Is this your first semester?

K: Yes we're still using the simple one. Yes I am a new comer and this is my first semester, Prof.


P: Good, it's a good enough for new student you can improve more.


Weak Point:

Still using a simple machine learning for classification (support vector machine)


Future Work:

In the future work, a system which can reveal the contents of the hidden messages will be extended.

Attachment 1:   첨부파일 4A-4.pdf(110.4KB)  
(Total:259 articles / page:1/26 )