Crossed-Time Delay Neural Network for Speaker Recognition

Time Delay Neural Network (TDNN) is a well-performing structure for deep neural network-based speaker recognition systems. In this paper we introduce novel structure, named Crossed-Time (CTDNN) to enhance the performance of current TDNN recognition. Inspired by multi-filters setting convolution layers from networks, set multiple time delay units with different context size at bottom layer and construct multilayer parallel network. The proposed CTDNN gives significant improvements over original on both verification identification tasks. It outperforms in VoxCeleb1 dataset experiment 2.6% absolute Equal Error Rate improvement. few shots condition, reaches 90.4% accuracy, which doubles accuracy TDNN. We also compare another new variant TDNN, Factorized-TDNN, shows that our model has 36% improvement under condition. Moreover, can handle training larger batch more efficiently hence, utilize calculation resources economically.