Structural Sparsification for Far-Field Speaker Recognition with Intel® Gna

Abstract

Recently, deep neural networks (DNN) have been widely used in speaker recognition area. In order to achieve fast response time and high accuracy, the requirements for hardware resources increase rapidly. However, as the speaker recognition application is often implemented on mobile devices, it is necessary to maintain a low computational cost while keeping high accuracy in far-field condition. In this paper, we apply structural sparsification on time-delay neural networks (TDNN) to remove redundant structures and accelerate the execution. On our targeted hardware, our model can remove 60% of parameters and only slightly increasing equal error rate (EER) by 0.18% while our structural sparse model can achieve more than 2× speedup.

Compact Speaker Embedding: lrx-Vector

“…Sharan et al [21] explore random projection for low-rank tensor factorization and describe the use on gene expression and EEG time series data. Zhang et al [22] apply structural sparsification on Time-Delay Neural Networks (TDNN) to remove redundant structures. Alternative approaches are subject to our further research, e.g., binary neural networks as successfully applied to natural language understanding [23].…”