Improving Time Delay Neural Network Based Speaker Recognition with Convolutional Block and Feature Aggregation Methods