Abstract
Existing speech perceptual hashing authentication algorithms generally encrypt the perceptual hash value to protect the security of the algorithms. However, under Kerckhoffs' principle, such algorithms can become transparent and fragile. Furthermore, security analysis based on Shannon unicity distance suggests that secret keys may be estimated when reused excessively. To address these vulnerabilities, we present a novel security authentication scheme for speech perceptual hashing based on the Fuzzy Commitment Scheme (FCS). In this scheme, a sender generates a commitment using a randomly selected BCH codeword and an extracted perceptual hash. The receiver then attempts to recover the codeword from the received speech and commitment. Experimental results demonstrate that the proposed scheme effectively prevents attackers from obtaining plaintext/ciphertext pairs without compromising the performance of the original hashing algorithm, thereby ensuring robust security for perceptual hashing.
1. Introduction
Perceptual hashing is a vital technique for identifying and authenticating multimedia content. Unlike traditional cryptographic hashes, which are highly sensitive to single-bit changes, perceptual hashes are designed to be robust against non-malicious distortions such as compression or noise. However, this inherent "fuzziness" introduces security challenges. Standard encryption of these hashes often fails to meet Kerckhoffs' Principle—where security should reside in the key, not the algorithm's secrecy—and is susceptible to key estimation via Shannon unicity distance analysis when keys are reused.
This paper proposes a solution by integrating a Fuzzy Commitment Scheme (FCS) with speech perceptual hashing. This approach decouples security from the hash itself and leverages error-correcting codes to bridge the gap between robust perceptual matching and hard cryptographic verification.
2. Methodology
The proposed system enhances existing speech hashing by incorporating an FCS layer. This layer uses Error Correcting Codes (ECC) to handle the minor variations between original and received speech signals while maintaining cryptographic integrity.
2.1 The Fuzzy Commitment Scheme (FCS)
The core of the security enhancement lies in the FCS, which uses a commitment $\delta$ to bind a random codeword to a perceptual hash.
As shown in Figure 1, the process is divided into two phases:
- Enrolment (Sender): A perceptual hash $H_r$ is extracted from the input message. A codeword $c$ is randomly selected from a BCH (Bose-Chaudhuri-Hocquenghem) code. The commitment $\delta$ is computed as $\delta = H_r \oplus c$. A cryptographic hash of the codeword, $h(c)$, is also generated. The pair $(h(c), \delta)$ is then transmitted.
- Verification (Receiver): A query hash $H_q$ is extracted from the received speech. The receiver computes a candidate codeword $c' = H_q \oplus \delta$. Using the BCH decoding function, any "noise" (differences between $H_r$ and $H_q$) is corrected. If the differences are within the error-correction threshold, the original $c$ is recovered. Finally, the system verifies the match by checking if $h(c') = h(c)$.
3. Speech Hashing Pipeline and Integration
The underlying perceptual hashing algorithm utilized in this study is based on Wavelet Packet Decomposition (WPD) and 2D-Discrete Cosine Transform (2D-DCT).
As illustrated in Figure 2, the feature extraction process involves pre-processing, WPD, blocking, and 2D-DCT. The resulting features are structured into a hash sequence. The integration with FCS (via the $Cmt$ and $Decmt$ functions) ensures that the transmitted hash information is cryptographically secure. The use of BCH codes $(n, k, t)$ allows for tunable robustness, where $t$ defines the maximum number of correctable errors, directly corresponding to the system's tolerance for speech distortion.
4. Experimental Results and Analysis
The proposed scheme was evaluated using Bit Error Rate (BER) and False Accept/Reject Rate (FAR/FRR) metrics. Key findings include:
- Maintained Performance: The FCS layer does not degrade the original algorithm's discriminative power. The system effectively distinguishes between legitimate speech and malicious tampering.
- Improved Reliability: The BCH code's error-correction capabilities slightly improve the BER distribution by mitigating minor transmission errors that would otherwise lead to authentication failure.
- Enhanced Security: By refreshing the random codeword $c$ for each session, the scheme prevents attackers from accumulating enough plaintext/ciphertext pairs to estimate a secret key, thus addressing the Shannon unicity distance problem.
5. Conclusion
Integrating a Fuzzy Commitment Scheme into speech perceptual hashing provides a robust security layer that satisfies modern cryptographic principles. By using error-correcting codes as a bridge, the system achieves high diffusion and protects against key estimation without sacrificing the robustness required for perceptual matching. Future work will focus on optimizing BCH parameters for diverse speech environments and exploring the generalizability of this framework to other perceptual hashing schemes.
Summary updated