A File Encryption Method Based On Voiceprint Recognition

Voiceprint recognition,as a biometric identification method with high usability,high accuracy,low cost and therefully a high user acceptance,is gradually being accepted in information security.In the context of protecting important personal documents,encryption algorithms have always been an important approach to the mainstream.The present invention combines voiceprint recognition with file encryption algorithms,proposing a file encryption method based on voiceprint recognition.Meanwhile,this system is optimized and improved for practical applications.The main contributions of this thesis are divided into the following three parts: 1.Propose an encryption method for extracting the key from the deep voiceprint feature x-vector embedding.The method extracts the 128-bit abstract from the x-vector embedding,as the key of the AES encryption algorithm,to encrypt and decrypt files.2.Design and implement the file encryption and decryption system based on voiceprint recognition.The system is divided into voiceprint recognition module,file encryption and decryption module,key generation module and storage management module,among which the core voiceprint recognition module.3.Aiming at two specific problems in the practical application of the x-vector model,improvement measures are proposed.One is to optimize the voice endpoint detection link in the audio frequency processing process,implementing different endpoint detection methods based on the signal-to-noise ratio to effectively improve the accuracy of endpoint detection in noisy environments;another is from the perspective of saving the time for extracting voiceprint embedding in practical applications,a feature optimization based on the x-vector method is proposed,extracting Fbank and input it into TDNN,which greatly reduces the time for users to extract voiceprint embedding.After testing,the system can achieve the designed functional indicators.The experimental results of the performance test section show that the improved x-vector feature optimization method in this paper has an EER of 4.21%,and can save more than 35% of the time to extract the voiceprint embedding.The improved endpoint detection method increases the accuracy of endpoint detection to 90% when the signal-to-noise ratio reaches more than 10 d B.In addition,the accuracy in actual use have reached the design specifications.