Locally-connected and convolutional neural networks for small footprint speaker recognition

This work compares the performance of deep LocallyConnected Networks (LCN) and Convolutional Neural Networks (CNN) for text-dependent speaker recognition. These topologies model the local time-frequency correlations of the speech signal better, using only a fraction of the number of parameters of a fully connected Deep Neural Network (DNN) used in previous works. We show that both a LCN and CNN can reduce the total model footprint to 30% of the original size compared to a baseline fully-connected DNN, with minimal impact in performance or latency. In addition, when matching parameters, the LCN improves speaker verification performance, as measured by equal error rate (EER), by 8% relative over the baseline without increasing model size or computation. Similarly, a CNN improves EER by 10% relative over the baseline for the same model size but with increased computation.