SpecWav-Attack: Leveraging Spectrogram Resizing and Wav2Vec 2.0 for Attacking Anonymized Speech

TL;DR

This paper introduces SpecWav-Attack, an adversarial approach that combines spectrogram resizing and Wav2Vec 2.0 to effectively attack anonymized speech systems, exposing their vulnerabilities and highlighting the need for enhanced defenses.

Contribution

It proposes a novel adversarial attack method leveraging spectrogram resizing and Wav2Vec 2.0, outperforming existing attacks on anonymized speech datasets.

Findings

Outperforms conventional attacks on librispeech datasets

Reveals vulnerabilities in anonymized speech systems

Highlights the need for stronger defenses

Abstract

This paper presents SpecWav-Attack, an adversarial model for detecting speakers in anonymized speech. It leverages Wav2Vec2 for feature extraction and incorporates spectrogram resizing and incremental training for improved performance. Evaluated on librispeech-dev and librispeech-test, SpecWav-Attack outperforms conventional attacks, revealing vulnerabilities in anonymized speech systems and emphasizing the need for stronger defenses, benchmarked against the ICASSP 2025 Attacker Challenge.