Phone Call Speaker Classification using Machine Learning on MFCC Features for Scam Detection
Yanisa Medrina Rahman, Yoanes Bandung · 2022
With the expansion of information and communication technology, human vulnerability to security threats is increasing. One of those threats is phone call-based fraud. Numerous studies on phone scam detection have been conducted. However, the proposed methods still have flaws, including identity spoofing, voice impersonation, and the risk of user privacy leakage. The lack of datasets is also a problem, particularly in the Indonesian language, where our research focuses. In this paper, we construct a phone call voice dataset from YouTube videos and conduct experiments on voice classification using machine learning and MFCC features in order to build a scam detection system that classifies speakers based on their voices and uses them as identifiers to build reputations. Support Vector Machine is the most accurate of the four machine learning classifiers we compare, achieving 94.46% accuracy using 68 MFCC features.