End-to-end DNN Based Speaker Recognition Inspired by i-vector and PLDA

TL;DR

This paper introduces an end-to-end DNN-based speaker verification system initialized to mimic traditional i-vector + PLDA systems, then fine-tuned to outperform the baseline on both long and short utterances.

Contribution

The authors propose a novel regularized training approach that improves end-to-end speaker verification performance by leveraging the i-vector + PLDA baseline as a starting point.

Findings

Outperforms i-vector + PLDA baseline on long utterances

Outperforms i-vector + PLDA baseline on short utterances

Mitigates overfitting in end-to-end systems

Abstract

Recently several end-to-end speaker verification systems based on deep neural networks (DNNs) have been proposed. These systems have been proven to be competitive for text-dependent tasks as well as for text-independent tasks with short utterances. However, for text-independent tasks with longer utterances, end-to-end systems are still outperformed by standard i-vector + PLDA systems. In this work, we develop an end-to-end speaker verification system that is initialized to mimic an i-vector + PLDA baseline. The system is then further trained in an end-to-end manner but regularized so that it does not deviate too far from the initial system. In this way we mitigate overfitting which normally limits the performance of end-to-end systems. The proposed system outperforms the i-vector + PLDA baseline on both long and short duration utterances.