Voice Command Fingerprinting: An in-Depth Examination of Challenges and Issues

Voice assistant systems have gained widespread adoption across diverse sectors, including electronics, healthcare, automotive, and retail. Spoken Language Understanding is essential for voice-text interaction systems. Security researchers have already identified vulnerabilities in encrypted network traffic, particularly in the context of traffic analysis, enabling the network-level adversaries to infer the websites visited by users. This security threat, common in traditional web communication, extends to Website Fingerprinting attacks in voice query communication, giving rise to the concept of Voice Command Traffic Fingerprinting. This paper analyses the features, challenges and linguistic properties of VCF(Voice Command Fingerprinting) considering various features. A detailed analysis reveals that the proposed models for voice command categories and factors are more capable to handle the issues. The results and methods in this paper pave the way for further research on the secured models in voice command processing in . The Python tools like Tensor-flow, Scikit-learn and specially Pyttsx3 have been used for generating and filtering datasets. An CTC-Attention-based hybrid model recurrent unit which effectively combine the two models. The proposed model achieves over 99.7% accuracy.