Disappeared Command: Spoofing Attack On Automatic Speech Recognition Systems with Sound Masking

TL;DR

This paper introduces a spoofing attack on automatic speech recognition systems using sound masking, highlighting vulnerabilities in deep learning-based ASR technology and its potential security risks.

Contribution

The paper presents a novel sound masking attack method that can deceive ASR systems, exposing security flaws in current deep learning speech recognition models.

Findings

Sound masking can effectively spoof ASR systems.

Deep learning ASR systems are vulnerable to subtle audio disturbances.

Potential security risks in voice-controlled applications.

Abstract

The development of deep learning technology has greatly promoted the performance improvement of automatic speech recognition (ASR) technology, which has demonstrated an ability comparable to human hearing in many tasks. Voice interfaces are becoming more and more widely used as input for many applications and smart devices. However, existing research has shown that DNN is easily disturbed by slight disturbances and makes false recognition, which is extremely dangerous for intelligent voice applications controlled by voice.