Adversarial Privacy Protection on Speech Enhancement

TL;DR

This paper introduces an adversarial approach to protect speech privacy by degrading speech enhancement systems, preventing malicious content extraction through targeted perturbations that significantly alter recognition results.

Contribution

It presents a novel adversarial method to disrupt speech enhancement systems, effectively safeguarding speech content from malicious extraction attacks.

Findings

Adversarial examples can erase or replace speech content during enhancement.

The method achieves an 89.0% word error rate in recognition of original speech.

Targeted attacks result in a low 33.75% WER with target speech content.

Abstract

Speech is easily leaked imperceptibly, such as being recorded by mobile phones in different situations. Private content in speech may be maliciously extracted through speech enhancement technology. Speech enhancement technology has developed rapidly along with deep neural networks (DNNs), but adversarial examples can cause DNNs to fail. In this work, we propose an adversarial method to degrade speech enhancement systems. Experimental results show that generated adversarial examples can erase most content information in original examples or replace it with target speech content through speech enhancement. The word error rate (WER) between an enhanced original example and enhanced adversarial example recognition result can reach 89.0%. WER of target attack between enhanced adversarial example and target example is low to 33.75% . Adversarial perturbation can bring the rate of change to the original example to more than 1.4430. This work can prevent the malicious extraction of speech.