The research article discusses a privacy-preserving approach for analyzing security in Voice over IP (VoIP) systems, focusing on the removal of silence portions in speech signals to enhance speaker detection amid encrypted communications. It utilizes a multi-layer perceptron and hidden Markov models for the classification of voiced and silent parts of speech, demonstrating improved endpoint detection and silence removal compared to conventional methods. The proposed method involves advanced traffic analysis attacks that can accurately detect speakers without needing simultaneous access to both ends of the communication.