Project DetailsInformation security and safety challenges are growing with technological advancement, and the need for high-quality, reliable authentication methods is becoming ever more urgent. The use of biometric authentication is expanding due to its ability to provide secure and convenient verification. In this project we are exploring the possibility of using the PPG signal – which reflects changes in blood vessel volume during cardiac cycles – as a means of biometric authentication. The PPG signal is influenced by an individual's physiological characteristics, thereby creating a unique biometric “fingerprint.” A significant advantage of this authentication method is that most smartwatches and fitness bands are already equipped with PPG sensors, and since this hardware is relatively inexpensive, obtaining the information is neither costly nor complicated. In addition, our project adopts advanced methods from the field of speech processing—a domain where efficient models for speaker verification based on dynamic and complex features have been developed. Since the uniqueness of both the PPG and speech signals is determined by physiological characteristics, and both exhibit quasi-periodic behavior, we show that applying speaker verification architectures can help capture subtle patterns in the PPG signal and improve recognition and authentication capabilities. In our approach, we take a PPG signal, preprocess it so that it resembles a speech signal in certain aspects, and then perform authentication using a machine learning model trained for speaker verification. With this method, the results we achieved surpass existing approaches in the field.
