Real-Time Classification of Real-Time Communications

Real-time communication (RTC) applications have become largely popular in the last decade with spread of broadband and mobile Internet access. Nowadays, these platforms are a fundamental means for connecting people supporting businesses that increasingly rely on forms remote work. In this context, it is paramount importance to operate at network level ensure adequate Quality Experience (QoE) users, appropriate traffic management policies essential prioritize RTC traffic. This turn requires be able identify streams type content they carry. paper, we propose machine learning-based application classify media generated by encapsulated Secure Real-Time Protocol (SRTP) flows real-time. Using carefully tuned features extracted from packet characteristics, train models into variety classes, including (audio/video), video quality, redundant streams. We validate our approach using over 62 hours multi-party meetings conducted two applications, namely Cisco Webex Teams Jitsi Meet. achieve an overall accuracy 96% 95% Jitsi, lightweight decision tree model makes decisions based solely 1 second real-time Our results show trained particular meeting software difficulty when used another one, although domain adaptation techniques facilitate transfer pre-trained models.