Short Speech Key Generation Technology Based on Deep Learning.

With the increasing popularity of biometric identity authentication in important key authentication applications, biometric key generation technology has attracted more and more attention. How to extract secure, stable and high-strength keys from biometrics has become a key research direction of biometric key generation technology. However, the proposed biometric key generation technology can not meet the security requirements as a key, that is, the key is easy to be cracked, resulting in unpredictable consequences. This paper presents a framework of voiceprint key generation based on deep neural network, which is mainly composed of voiceprint feature extraction model and fuzzy extraction model. The voiceprint feature extraction model is based on the deep neural network, and the sigmoid layer is added at the end of the neural network. This operation makes the features extracted by the neural network have high accuracy and low error recognition rate after binary quantization. The voiceprint database used in this paper is aishell-wakeup-1 wakeup word database. The voiceprint key generation model proposed in this paper generates the key on the voiceprint database with the generation intensity of more than 1000bit, the accuracy rate of more than 93%, and the error recognition rate of less than 0.001%. These data can prove that the voiceprint key generation model proposed in this paper can meet the needs of users for generation strength and security.