Security Analysis of Speech Perceptual Hashing Authentication Algorithm

Qiuyu Zhang, Ren Zhan-wei, Xing Peng-fei, Yibo Huang, Shuang YuInternational Journal of Security and Its Applications 2016 · DOI

Abstract

Speech perceptual hashing authentication algorithm is an efficient method for content integrity authentication and identity authentication. But the algorithm becomes transparent under the principle of Kerckhoffs which makes the algorithm unsafe. In this paper, the algorithm is encrypted under the principle of Kerckhoffs to protect the security of the algorithm. Then the specific link that won't affect the performance of the algorithm is identified through the experiments. Next, this paper analyzes the security of the encrypted algorithm based on the concept of Shannon unicity distance. The unicity distance is figured out finally based on the experiments. That is to say, the algorithm loses its security even though the algorithm has been encrypted. Under this circumstances, the most important thing is to continue keeping the safety of the algorithm. Therefore, this paper proposed an efficient random secret key method to guarantee the safety of the algorithm after the unicity distance is figured out.

Introduction

Speech perceptual hashing is a "fuzzy" hashing technique used to authenticate audio while ignoring harmless edits like compression or volume adjustment. Unlike standard cryptographic hashes (such as SHA-256), where changing a single bit results in a completely different output, perceptual hashes remain similar if the audio undergoes content-preserving changes. They only change drastically if the actual meaning or content of the speech is tampered with.

A significant challenge in this field is Kerckhoffs’s Principle: a secure system should remain safe even if the attacker knows exactly how the algorithm works, provided the key remains secret. Most existing speech hashing algorithms are "transparent," meaning an attacker who knows the algorithm can forge hashes or bypass authentication. While adding encryption is a common response, there is little analysis on how much security that encryption provides or how it impacts the "perceptual" performance of the hash.

Methodology and Encryption Placement

The researchers built upon an efficient hashing algorithm based on Wavelet Packet Decomposition (WPD). A primary goal was to identify the optimal stage to insert encryption within the hashing pipeline without destroying the algorithm's robustness.

Figure 2: Flowchart showing the various stages where encryption was tested within the feature extraction and matching process.

Through simulation, the authors tested two primary locations: at the internal feature extraction stage and at the final hash generation stage. They discovered that encrypting internal features makes the hash too sensitive to harmless edits, effectively breaking its perceptual utility. However, encrypting the final binary hash sequence preserved the necessary performance characteristics.

Security Modeling via Unicity Distance

To quantify security, the authors applied Shannon’s concept of unicity distance—the amount of ciphertext required to uniquely determine a secret key. They modeled the encryption as a linear algebra problem: $V = KI$, where $I$ is the plaintext hash, $V$ is the encrypted ciphertext, and $K$ is the secret key matrix.

Using the least squares method, they simulated an attacker trying to estimate the secret key $K$ by intercepting pairs of plaintext and ciphertext hashes. For a hash of length $m=256$, the experiments showed that once an attacker intercepts 256 pairs, the secret key can be solved via matrix inversion, reducing the security distance to zero.

Performance and Results

A critical finding was that adding security did not come at the cost of accuracy. The authors compared the False Accept Rate (FAR) and False Reject Rate (FRR) curves to ensure the "perceptual" nature of the hash remained intact.

Figure 5: Comparison of FAR-FRR curves before (a) and after (b) encryption, demonstrating that performance remains stable.

As seen in Figure 5, the curves are nearly identical, proving that encryption can be added without sacrificing the ability to distinguish between legitimate edits and malicious tampering. To address the vulnerability found in the unicity distance analysis, the authors proposed a proactive random secret key method. By rotating the secret key before the number of uses reaches the unicity distance (the 256-pair limit), the algorithm stays ahead of potential attackers.

Conclusion

This research demonstrates that while static encryption of speech hashes is vulnerable to mathematical estimation over time, dynamic key rotation provides a robust solution. By identifying the correct encryption link and calculating the unicity distance, developers can maintain the security of perceptual hashing systems under the rigorous standards of Kerckhoffs’s Principle.

Summary updated

Abstract

Speech perceptual hashing authentication algorithm is an efficient method for content integrity authentication and identity authentication. But the algorithm becomes transparent under the principle of Kerckhoffs which makes the algorithm unsafe. In this paper, the algorithm is encrypted under the principle of Kerckhoffs to protect the security of the algorithm. Then the specific link that won't affect the performance of the algorithm is identified through the experiments. Next, this paper analyzes the security of the encrypted algorithm based on the concept of Shannon unicity distance. The unicity distance is figured out finally based on the experiments. That is to say, the algorithm loses its security even though the algorithm has been encrypted. Under this circumstances, the most important thing is to continue keeping the safety of the algorithm. Therefore, this paper proposed an efficient random secret key method to guarantee the safety of the algorithm after the unicity distance is figured out.
BibTeX
@article{zhang2016security,
  author = {Qiuyu Zhang and Ren Zhan-wei and Xing Peng-fei and Yibo Huang and Shuang Yu},
  title = {{Security Analysis of Speech Perceptual Hashing Authentication Algorithm}},
  journal = {International Journal of Security and Its Applications},
  year = {2016},
  doi = {10.14257/ijsia.2016.10.1.11},
}
Figure