[Paper Review] Real-Time Lightweight Chaotic Encryption for 5G IoT Enabled Lip-Reading Driven Secure Hearing-Aid

Ahsan Adeel, Jawad Ahmad+1 more|arXiv (Cornell University)|Sep 13, 2018

Speech and Audio Processing17 references4 citations

TL;DR

This paper proposes a 5G IoT-enabled, real-time lightweight audio-visual hearing-aid system that leverages lip-reading-driven deep learning for speech enhancement and a novel chaotic encryption scheme for secure communication. The framework achieves sub-5ms round-trip latency, outperforms audio-only methods in low SNR environments (e.g., -12dB SNR), and demonstrates strong security with 99.95% NPCR and 33.39% UACI.

ABSTRACT

Existing audio-only hearing-aids are known to perform poorly in noisy situations where overwhelming noise is present. Next-generation audio-visual (lip-reading driven) hearing-aids stand as a major enabler to realise more intelligible audio. However, high data rate, low latency, low computational complexity, and privacy are some of the major bottlenecks to the successful deployment of such advanced hearing aids. To address these challenges, we envision an integration of 5G Cloud-Radio Access Network, Internet of Things (IoT), and strong privacy algorithms to fully benefit from the possibilities these technologies have to offer. The envisioned 5G IoT enabled secure audio-visual (AV) hearing-aid transmits the encrypted compressed AV information and receives encrypted enhanced reconstructed speech in real-time which fully addresses cybersecurity attacks such as location privacy and eavesdropping. For security implementation, a real-time lightweight AV encryption is utilized. For speech enhancement, the received AV information in the cloud is used to filter noisy audio using both deep learning and analytical acoustic modelling (filtering based approach). To offload the computational complexity and real-time optimization issues, the framework runs deep learning and big data optimization processes in the background on the cloud. Specifically, in this work, three key contributions are reported: (1) 5G IoT enabled secure audio-visual hearing-aid framework that aims to achieve a round-trip latency up to 5ms with 100 Mbps datarate (2) Real-time lightweight audio-visual encryption (3) Lip-reading driven deep learning approach for speech enhancement in the cloud. The critical analysis in terms of both speech enhancement and AV encryption demonstrate the potential of the envisioned technology in acquiring high-quality speech reconstruction and secure mobile AV hearing aid communication.

Motivation & Objective

• Address the limitations of existing audio-only hearing aids in noisy environments, where speech intelligibility remains poor despite amplification.
• Overcome challenges in real-time AV hearing-aid deployment, including high data rate, low latency, low computational complexity, and strong privacy requirements.
• Integrate 5G Cloud-Radio Access Network (C-RAN) and IoT to enable secure, low-latency transmission of compressed audio-visual data.
• Develop a lightweight, real-time encryption scheme resistant to common cyberattacks such as eavesdropping and location tracking.
• Enable cloud-based deep learning and acoustic modeling to offload computational complexity while maintaining real-time performance.

Proposed method

• Employ a piece-wise linear chaotic map (PWLCM) and Chebyshev map to generate pseudorandom sequences for stream cipher-based audio-visual encryption.
• Design a novel substitution box (S-Box) using chaotic maps and secure hash functions to enhance confusion and diffusion in the encryption process.
• Utilize a lip-reading-driven deep learning model (LSTM-based) in the cloud to reconstruct enhanced speech from compressed audio-visual inputs.
• Apply analytical acoustic modeling (filtering-based approach) in parallel with deep learning to improve speech enhancement robustness.
• Offload heavy computation (deep learning and big data optimization) to the cloud via 5G-CRAN to meet real-time constraints.
• Integrate end-to-end encryption: original AV data is compressed and encrypted at the hearing aid before transmission, and decrypted only after enhanced speech is received.

Experimental results

Research questions

• RQ1 Can a 5G IoT-enabled audio-visual hearing-aid achieve real-time performance with round-trip latency under 5ms and data rate of 100 Mbps?
• RQ2 How effective is the proposed lightweight chaotic encryption scheme in securing AV data against common cyberattacks while maintaining low computational overhead?
• RQ3 To what extent does lip-reading-driven deep learning improve speech enhancement in low SNR environments compared to audio-only methods?
• RQ4 What is the security strength of the proposed encryption scheme in terms of key space, randomness, and resistance to differential attacks?
• RQ5 How does the proposed system perform in real-world noisy scenarios such as cafes, streets, and public transport?

Key findings

• The proposed 5G IoT-enabled AV hearing-aid framework achieved a round-trip latency of less than 5ms at a 100 Mbps data rate, meeting real-time communication requirements.
• The audio encryption scheme achieved a Number of Pixel Change Rate (NPCR) of 99.95% and Unified Average Change Intensity (UACI) of 33.39%, indicating high sensitivity to input changes and strong resistance to differential attacks.
• The correlation coefficient of the encrypted signal was 0.00022, demonstrating near-zero correlation between adjacent samples and effective information hiding.
• The key length of the proposed scheme was estimated at 10^45, far exceeding the minimum threshold of 2^100, ensuring strong resistance against brute-force attacks.
• In low SNR conditions (-12dB, -6dB, -3dB), the proposed EVWF (Enhanced Visual-aided Fusion) speech enhancement method significantly outperformed audio-only benchmarks (SS and LMMSE), with PESQ scores indicating superior speech quality.
• Subjective listening tests using MOS (Mean Opinion Score) showed that the proposed AV method achieved a score of 4.2 at -12dB SNR, outperforming audio-only methods, and maintained comparable performance to audio-only approaches at high SNRs.