Audio Adversarial Examples: Attacks Using Vocal Masks

TL;DR

This paper introduces a novel audio adversarial attack method using vocal masks that successfully fools state-of-the-art speech-to-text systems while remaining recognizable to humans, highlighting vulnerabilities in current ASR models.

Contribution

The authors propose a new vocal mask-based adversarial attack that effectively deceives multiple leading speech-to-text systems without hindering human comprehension.

Findings

Adversarial examples fool SOTA STT systems

Humans can still accurately transcribe masked audio

The attack reveals vulnerabilities in current ASR models

Abstract

We construct audio adversarial examples on automatic Speech-To-Text systems . Given any audio waveform, we produce an another by overlaying an audio vocal mask generated from the original audio. We apply our audio adversarial attack to five SOTA STT systems: DeepSpeech, Julius, Kaldi, wav2letter@anywhere and CMUSphinx. In addition, we engaged human annotators to transcribe the adversarial audio. Our experiments show that these adversarial examples fool State-Of-The-Art Speech-To-Text systems, yet humans are able to consistently pick out the speech. The feasibility of this attack introduces a new domain to study machine and human perception of speech.