STC Antispoofing Systems for the ASVspoof2019 Challenge

View on arXiv ← Back to list

Authors: Galina Lavrentyeva, Sergey Novoselov, Andzhukaev Tseren, Marina Volkova, Artem Gorlanov, Alexandr Kozlov

Published: 2019-04-11 08:37:43+00:00

AI Summary

This paper presents the Speech Technology Center's (STC) anti-spoofing systems for the ASVspoof 2019 challenge. The systems, based on an enhanced Light CNN architecture with angular margin-based softmax activation, achieved low equal error rates (EERs) of 1.86% and 0.54% in logical and physical access scenarios, respectively.

Abstract

This paper describes the Speech Technology Center (STC) antispoofing systems submitted to the ASVspoof 2019 challenge. The ASVspoof2019 is the extended version of the previous challenges and includes 2 evaluation conditions: logical access use-case scenario with speech synthesis and voice conversion attack types and physical access use-case scenario with replay attacks. During the challenge we developed anti-spoofing solutions for both scenarios. The proposed systems are implemented using deep learning approach and are based on different types of acoustic features. We enhanced Light CNN architecture previously considered by the authors for replay attacks detection and which performed high spoofing detection quality during the ASVspoof2017 challenge. In particular here we investigate the efficiency of angular margin based softmax activation for training robust deep Light CNN classifier to solve the mentioned-above tasks. Submitted systems achieved EER of 1.86% in logical access scenario and 0.54% in physical access scenario on the evaluation part of the Challenge corpora. High performance obtained for the unknown types of spoofing attacks demonstrates the stability of the offered approach in both evaluation conditions.

Key findings

The proposed systems achieved excellent performance on the ASVspoof 2019 challenge, demonstrating the effectiveness of the enhanced Light CNN architecture. The angular margin-based softmax activation and batch normalization improved the model's performance and stability. The results highlight the potential of deep learning for robust spoofing detection, though the reliance on simulated data raises concerns about real-world applicability.

Approach

The authors enhanced a Light CNN architecture by incorporating angular margin-based softmax activation and batch normalization. This improved the robustness and convergence speed during training for spoofing detection using various acoustic features (LFCC, CQT, FFT, DCT). A score-level fusion of multiple systems was used for final submission.

Datasets

ASVspoof 2019 dataset

Model(s)

Enhanced Light CNN with angular margin-based softmax activation

Author countries

Russia

← Previous