Robust Training for Speaker Verification against Noisy Labels

TL;DR

This paper introduces a two-stage training approach for speaker verification models that effectively filters out noisy labels, improving robustness against label errors in large datasets.

Contribution

The proposed method uses a novel OR-Gate with top-k mechanism to iteratively select clean data, enhancing model performance in noisy label scenarios.

Findings

Achieved high accuracy on VoxCeleb datasets with added noise.

Effectively filters noisy labels through iterative training.

Demonstrates robustness of the method against various noise levels.

Abstract

The deep learning models used for speaker verification rely heavily on large amounts of data and correct labeling. However, noisy (incorrect) labels often occur, which degrades the performance of the system. In this paper, we propose a novel two-stage learning method to filter out noisy labels from speaker datasets. Since a DNN will first fit data with clean labels, we first train the model with all data for several epochs. Then, based on this model, the model predictions are compared with the labels using our proposed the OR-Gate with top-k mechanism to select the data with clean labels and the selected data is used to train the model. This process is iterated until the training is completed. We have demonstrated the effectiveness of this method in filtering noisy labels through extensive experiments and have achieved excellent performance on the VoxCeleb (1 and 2) with different added noise rates.