Research On Encrypted Speech Retrieval Method And Index Scheme Based On Deep Hashing

With the development of cloud storage technology and the explosion of multimedia data,research on cloud data retrieval methods has become a hot topic.Speech has a special semantic function,which contains rich semantic information and content perception information,it plays a vital role in court evidence,military secrets and other applications.In order to protect these important private data,encryption operations are indispensable,but encryption operations will bring certain challenges to the retrieval tasks.Therefore,the study of cloud encrypted speech retrieval schemes has great research value.The thesis mainly uses deep neural network model model,hash function construction,speech signal processing and other technologies to conduct research on key technologies such as deep semantic feature extraction,deep hash construction,speech classification,index scheme construction and so on.The main research work is as follows:1.In order to solve the problems of poor manual feature semantics,low retrieval accuracy and retrieval efficiency in existing speech feature extraction methods,an encrypted speech retrieval method based on CNN and deep hashing is proposed.Firstly,the Rossler chaotic map encryption method is used to construct the encrypted speech library.Then the second feature extraction method is used to extract the spectrogram features and high-level semantic features of the speech,and the deep hash binary code is generated by the designed CNN network model for the retrieval task.At the same time,the batch normalization method(BN)is introduced to improve the robustness and generalization ability of the model.The experimental results show that the deep hash binary code constructed by this method has strong discriminability and robustness,and still has high retrieval accuracy and retrieval efficiency under various content preserving operations.Meanwhile,the adopted speech encryption method has a higher key space,which can effectively resist the exhaustive attack.2.In order to solve the problems of low classification accuracy and complex classification model construction of traditional classification methods,and further improve the semantics of speech features and retrieval efficiency,an encrypted speech classification retrieval method based on CNN/CRNN and deep hashing is proposed.Firstly,the Rossler chaotic map encryption method is used to construct the encrypted speech library.Secondly,the Log-Mel spectrogram features of speech are extracted,and the semantic feature hash codes and classification results are generated by the designed CNN/CRNN model and the constructed hash function.Finally,the "two-stage" classification retrieval strategy and normalized Hamming distance algorithm are used to achieve matching retrieval.Experimental results show that the proposed CNN/CRNN coding model has excellent feature learning performance,and the retrieval method has well recall rate,precision rate and retrieval efficiency.3.In order to solve the problems of complex index data structure model and low search efficiency in traditional index methods,and to further improve retrieval efficiency and scalability of index structure,An index structure based on multi-hash tables is designed and applied to the encrypted speech retrieval system.Firstly,the idea of "hash code segmentation" is adopted on the semantic feature hash,according to the segmentation,several different hash index tables are generated.Then,the corresponding multiple sub-hash index tables are established in the hash table of each segments,so as to realize the parallel retrieval of multiple hash tables.Finally,a search method that gradually increases the search radius is adopted to complete different retrieval tasks according to different retrieval requirements.Experimental results show that the index structure can meet different data volume environments and retrieval requirements,has good scalability,and has high retrieval accuracy and retrieval efficiency.