An Improved Deep Neural Network for Modeling Speaker Characteristics at Different Temporal Scales

Abstract

This paper presents an improved deep embedding learning method based on convolutional neural network (CNN) for text-independent speaker verification. Two improvements are proposed for x-vector embedding learning: (1) Multiscale convolution (MSCNN) is adopted in frame-level layers to capture complementary speaker information in different receptive fields.(2) A Baum-Welch statistics attention (BWSA) mechanism is applied in pooling-layer, which can integrate more useful long-term speaker characteristics in the temporal pooling layer. Experiments are carried out on the NIST SRE16 evaluation set. The results demonstrate the effectiveness of MSCNN and show the proposed BWSA can further improve the performance of the DNN embedding system.

Speaker Recognition Based on Deep Learning: An Overview

“…• Single-head Baum-Welch statistics attention mechanism based statistics pooling [116]: To overcome the weakness of ( 12) which cannot fully mine the inner relationship between an utterance and its frames, [116] integrated the Baum-Welch statistics into the attention mechanism:…”

“…The key matrix K is calculated from the Baum-Welch statistics. Specifically, [116] first calculates the normalized first order statistics f c from the cth component of a GMM-UBM model Ω (see ( 5)), and then conducts the following nonlinear transform:…”