There is always the risk that the audio recordings of the conversations will be sent through encrypted public networks and then kept and processed in the cloud by an external service provider. Voice-based interactive technologies are evolving towards an always-listening mode, inspired by the idea of continuous user involvement as opposed to being reliant on traditional push-to-talk capability. Rather than sharing the real voice model, this technological task is carried out through the construction of the chaff model that will be given to the server. Using the Gaussian mixture model, the chaff model of the voice signal is initially established. The created chaff model will be fed into the RSA model to generate the secret key pair. The resulting key pair will be used to encode the voice chaff model, which will then be shared with the server. The actual voice signal will be extracted at the server end using the appropriate keys, and it will then be authenticated using the convolutional neural network.

Chaff Model, Gaussian Mixture Model, Neural Network, RSA and Voice Model