Spoofing Detection in Speaker Verification System

Abstract

Speech is the most common form of communication for humans. With the advances in artificial intelligence, humans can communicate to machines using speech, mainly used for biometrics. When there is an attempt to bypass the system with unethical means of using any fake form of speech it is known as speech spoofing. With advances in speech technology, it is the need of the hour to develop efficient and trustworthy speech recognizing and verifying systems. To develop a good spoofing detection system the three aspects of machine and deep learning are taken into consideration for experimental purposes. First is the feature extraction where a systematic approach is being followed to extract clean ad vital information from the raw speech data. With the use of that, an approach is proposed to develop a system to detect replay spoofing attempts. The second aspect is feature selection where a subset of important features is obtained from the extracted features to develop a spoofing detection system. The third and last aspect of transfer learning is taken into consideration for developing a system to detect spoofing attempts.