A secure marine mammal audio retrieval algorithm based on deep biohashing

Marine mammals are crucial subjects for marine audio studies, particularly in the development of retrieval algorithms for encrypted marine mammal audio. To address the issues of low retrieval performance and insufficient expression of audio features, and to ensure the security of existing audio retrieval algorithms, this paper proposes a secure marine mammal audio retrieval algorithm based on deep biohashing. The server terminal utilizes the BiGRU-Residual-Multi-Head-Self-Attention (BiGRU-RMHSA) autoencoder model to extract the deep features of dynamic Gammatone Cepstral Coefficients (GTCC), which improves robustness and discrimination of retrieval algorithm. Then, the classified deep features and the corresponding orthogonal matrix construct the key-controlled biosafety template to reduce the leakage of deep features. Finally, the template is quantified to obtain the biohashing. Meanwhile, the chaotic DNA encryption scheme encrypts audio to ensure audio security. The experiments show that the retrieval scheme has higher precision and recall for different content preserving operations. And the proposed encryption scheme reduces the audio correlation and its key space is 10 80 , ensuring the resistance of encrypted audio to statistical attacks. Furthermore, the biosafety template has better revocability, security and diversity. • The designed BiGRU-RMHSA autoencoder model extracts deep features of dynamic GTCC, which effectively shows better robustness and discrimination, further improving both recall and precision of retrieval system. • The paper designs the key-controlled biosafety template, updating the key of template to generate a new template when the original template is leaked. So the biosafety template has better diversity and revocability, which guarantees the security of the audio feature. • The chaotic DNA encryption algorithm introduces the 5D-CNN chaotic sequence into DNA encryption algorithms, which reduces the correlation of audio information and contains a larger key space, further effectively ensuring the privacy of audio in the cloud terminal.