Assessment of a VoIP steganalysis method based on statistical analysis and deep neural network

Area of Science:

• Cybersecurity
• Digital Signal Processing
• Artificial Intelligence

Background:

• Voice over Internet Protocol (VoIP) is increasingly used for covert steganography.
• Existing steganalysis methods struggle with high accuracy and efficiency.
• Integrating signal processing and machine learning offers enhanced detection capabilities.

Purpose of the Study:

• To propose a novel hybrid method combining speech signal processing and AI for VoIP steganography detection.
• To evaluate the proposed method's accuracy and computational efficiency against known steganography techniques.
• To demonstrate the model's effectiveness for real-time steganalysis applications.

Main Methods:

• Applied data preprocessing to G.729 compressed audio signals to extract intra-frame features and inter-frame correlations.
• Utilized a deep learning network for training to differentiate between cover and stego data.
• Assessed the hybrid method against Quantization Index Modulation (QIM), Pitch Modulation Steganography (PMS), and Heterogeneous Parallel Steganography (HPS).

Main Results:

• The hybrid method demonstrated significant improvements in detection accuracy and computational efficiency.
• Achieved high accuracy rates: 98.85% for QIM, 96.94% for PMS, and 91.90% for HPS.
• Response time for steganalysis testing was under 5ms for 1000 ms audio files, indicating high speed.

Conclusions:

• The proposed hybrid AI and signal processing technique offers superior accuracy and efficiency for VoIP steganography detection compared to conventional methods.
• The model's high-speed performance makes it suitable for real-time steganalysis.
• This research contributes a robust solution to counter security threats posed by steganography in VoIP communications.