Abstract: Currently, most speaker recognition systems make a decision by calculating the similarity between enrollment and test embeddings extracted with convolutional neural networks. However, for ...