Investigating Traffic of Smart Speakers and IoT Devices
Davide Caputo, Luca Verderame, Alessio Merlo, Luca Caviglione · 2020
This chapter investigates the feasibility of adopting machine-learning-based techniques to breach the privacy of users interacting with smart speakers or voice assistants. The first experiment aimed at investigating whether it is possible to identify if the microphone of the smart speaker or the device hosting the Intelligent Virtual Assistant is turned on or off. Smart speakers and voice-based virtual assistants are important building blocks of modern smart homes. For instance, they can be used to retrieve information, interact with other devices, and command a wide range of Internet of Things nodes. In other words, the larger the window, the more packets are contained, thus less information is provided to the machine learning algorithm. Support Vector Machine (SVM) is a group of supervised learning models for solving both regression or classification problems. SVM builds non-probabilistic binary classifiers whose purpose is the search for the optimal hyperplane of separation between the several possible classes within the feature space.