Time Domain Adversarial Voice Conversion for ADD 2022

TL;DR

This paper presents a time domain adversarial voice conversion system for the ADD 2022 challenge, capable of generating convincing fake speech that can deceive anti-spoofing detectors while maintaining good audio quality.

Contribution

The authors introduce a novel time domain post-processing step for voice conversion that enhances deception ability against anti-spoofing detectors.

Findings

System ranks top in ADD 2022 Track 3.1

Demonstrates strong adversarial ability against anti-spoofing detectors

Maintains acceptable audio quality and speaker similarity

Abstract

In this paper, we describe our speech generation system for the first Audio Deep Synthesis Detection Challenge (ADD 2022). Firstly, we build an any-to-many voice conversion (VC) system to convert source speech with arbitrary language content into the target speaker%u2019s fake speech. Then the converted speech generated from VC is post-processed in the time domain to improve the deception ability. The experimental results show that our system has adversarial ability against anti-spoofing detectors with a little compromise in audio quality and speaker similarity. This system ranks top in Track 3.1 in the ADD 2022, showing that our method could also gain good generalization ability against different detectors.