Hybrid ChaCha20 Framework Encrypts High Bit-Depth Audio with Near-Lossless Decryption

Researchers Lakhichand Patil and Kalpesh A. Popat have introduced an efficient hybrid audio encryption scheme that integrates the ChaCha20 stream cipher into sine and cosine non-linear transformations to secure high bit-depth sound files. Published in the journal Multimedia Tools and Applications, the framework is designed to process large audio recordings in uncompressed WAV format across bit depths of 16 bits, 24 bits, and 32 bits. The proposed architecture applies multi-stage diffusion and confusion across each individual sample of the audio signal while operating at very low computational costs. Designed for practical implementation, the authors position the scheme as a viable candidate for multimedia security in real-time and embedded systems, pointing to operational domains such as digital forensic analysis, broadcast protection, and the archival of professional audio material.

To establish the operational validity of the pipeline, the researchers conducted an experimental evaluation across a corpus of 1,500 uncompressed PCM audio files in WAV format. The testing dataset was balanced evenly, incorporating exactly 500 files for each of the three bit-depth categories of 16 bits, 24 bits, and 32 bits. The audio recordings were compiled from the MUSAN dataset and the Free Music Archive through stratified sampling across speech, music, and ambient sound environments. The experimental framework evaluated performance across two distinct comparative benchmarks: obfuscation quality, which assesses the difference between the original plaintext audio and the encrypted ciphertext, and reconstruction fidelity, which measures the accuracy of decrypted audio relative to the original plaintext input. These assessments relied on standard metrics including Peak Signal-to-Noise Ratio, Signal-to-Noise Ratio, Mean Squared Error, and the correlation coefficient.

In reconstruction benchmarks, the system demonstrated near-lossless recovery across every tested bit depth. The linear correlation between the original plaintext audio and the decrypted audio was exactly 1.000000 with zero variance across all 1,500 evaluated files. Across the entire test corpus, the maximum absolute sample error observed anywhere between the plaintext and the recovered signal was restricted to a single least-significant bit. A total of 337 files, representing 22.5 percent of the 1,500 files in the corpus, achieved completely bit-exact reconstruction without any sample discrepancies. For the remaining audio files where single least-significant bit errors occurred, the mean reconstruction Peak Signal-to-Noise Ratio was measured at 125.86 dB for 16-bit files, 174.57 dB for 24-bit files, and 222.67 dB for 32-bit files, confirming near-perfect acoustic reproduction upon decryption.

Conversely, evaluations within the obfuscation domain confirmed that the encrypted ciphertext exhibits properties indistinguishable from random noise. The framework yielded an average obfuscation Peak Signal-to-Noise Ratio of 4.41 dB with an uncertainty of plus or minus 0.47 dB, alongside an obfuscation Signal-to-Noise Ratio of −14.70 dB with an uncertainty of plus or minus 7.16 dB. The correlation coefficient measured between the ciphertext and original plaintext audio was −2.4 × 10⁻⁵ with an uncertainty of plus or minus 0.0024. The encrypted output demonstrated high randomness, recording a ciphertext byte entropy of 7.9997 bits out of an ideal maximum of 8.0000 bits. Additionally, the obfuscation Mean Squared Error scaled consistently with bit depth as anticipated by the researchers, increasing from 3.94 × 10⁸ at 16 bits to 1.67 × 10¹⁸ at 32 bits.

Operational testing on an otherwise idle machine demonstrated substantial computational efficiency and low algorithmic delay. The implementation achieved a processing throughput ranging from 11.5 million to 12.2 million samples per second. This processing speed translates to a real-time factor of 0.0044, indicating that encryption operates approximately 227 times faster than normal audio playback. The architecture organizes audio processing into keyed permutation blocks of 4,096 samples, introducing an algorithmic latency of 256 milliseconds at an audio sampling rate of 16 kHz and 93 milliseconds at 44.1 kHz. These timing metrics demonstrate that the hybrid framework is suitable for real-time live streaming environments as well as the encryption of stationary audio files stored at rest.

The cryptographic confidentiality of the proposed method rests primarily on ChaCha20 and its 256-bit key space, representing 2²⁵⁶ possible key configurations and comfortably exceeding the NIST-recommended minimum standard of 2¹²⁸. Key-sensitivity differential testing using single-bit key perturbations produced a number of pixels change rate of approximately 99.61 percent and a unified average changing intensity of approximately 33.46 percent, matching the expected avalanche dynamics produced by a 20-round ChaCha20 keystream. Furthermore, all eleven applicable tests from the NIST SP 800-22 randomness test suite were conducted, and all eleven tests successfully passed at a significance level of α = 0.01 under the testing suite’s established proportion criterion.

The investigators explicitly clarified the functional scope and contribution of the non-linear transformations and scramble and substitute stages. These components provide structural obfuscation of the audio representation, but the authors emphasize that they do not claim to add cryptographic strength beyond what ChaCha20 provides alone. A controlled comparative analysis revealed that ChaCha20 deployed by itself without transformation stages produces ciphertext statistically indistinguishable from the full hybrid pipeline across entropy, correlation, silence fraction, and spectral flatness. Consequently, under correct cryptographic deployment, the non-linear transformations offer no measurable obfuscation beyond the base stream cipher. However, the transformations offer substantial benefits during stream-cipher nonce reuse, a common operational misuse scenario. Under simulated nonce reuse, the transformations materially suppress data leakage: the correlation between the leaked difference signal and the true plaintext difference decreases from 1.0000 to 0.0031, its temporal correlation drops from 0.4333 to 0.0448, and its entropy increases from 6.3148 to 7.9160 bits per byte. The practical role of the non-linear stages is thus characterized as providing graceful degradation under cipher misuse rather than supplementary strength under proper operation.

Subject of Research: Hybrid ChaCha20 and non-linear transformation framework for high bit-depth audio encryption

Article Title: A hybrid audio encryption framework using ChaCha20 and non-linear transformations for high bit-depth audio with NIST randomness validation

Article References: Patil, L., & Popat, K. A. (2026). A hybrid audio encryption framework using ChaCha20 and non-linear transformations for high bit-depth audio with NIST randomness validation. Multimedia Tools and Applications, 85(11), Article 838. https://doi.org/10.1007/s11042-026-22011-0

Image Credits: AI Generated

DOI: 10.1007/s11042-026-22011-0

Keywords: audio encryption, ChaCha20, cryptography, NIST SP 800-22, signal processing, multimedia security, WAV format, stream ciphers, hybrid, audio, encryption, framework