Rapid model retraining for a new attack vector

A non-transitory computer-readable storage medium for use in conjunction with a computer system, the computer-readable storage medium configured to store program instructions that, when executed by the computer system, causes the computer system to perform one or more operations comprising:

• receiving information specifying or associated with a new attack vector corresponding to fake audio content;
• generating a synthetic training dataset based at least in part on the new attack vector, wherein the synthetic training dataset comprises synthetic audio content with fake audio content;
• accessing a predetermined neural network configured to classify real audio content and fake audio content, wherein the predetermined neural network was training using a second training dataset that excluded the synthetic audio content corresponding to the new attack vector and that has a larger size than a size of the synthetic training dataset, and wherein the predetermined neural network comprises convolutional blocks, arranged sequentially, followed by a softmax layer;
• and
• training a neural network based at least in part on the synthetic training dataset and the predetermined neural network, wherein the training of the neural network comprises modifying predetermined weights associated with the predetermined neural network, and wherein a training time for training the neural network is less than a training time for training the predetermined neural network.

A method for training a neural network, wherein the method comprises:

• receiving information specifying or associated with a new attack vector corresponding to fake audio content;
• generating a synthetic training dataset based at least in part on the new attack vector, wherein the synthetic training dataset comprises synthetic audio content with fake audio content;
• accessing a predetermined neural network configured to classify real audio content and fake audio content, wherein the predetermined neural network was training using a second training dataset that excluded the synthetic audio content corresponding to the new attack vector and that has a larger size than a size of the synthetic training dataset, and wherein the predetermined neural network comprises convolutional blocks, arranged sequentially, followed by a softmax layer;
• and
• training the neural network based at least in part on the synthetic training dataset and the predetermined neural network, wherein the training of the neural network comprises modifying predetermined weights associated with the predetermined neural network, and wherein a training time for training the neural network is less than a training time for training the predetermined neural network.