We construct targeted audio adversarial examples on automatic speech
recognition. Given any audio waveform, we can produce another that is over
99.9% similar, but transcribes as any phrase we choose (recognizing up to 50
characters per second of audio). We apply our white-box iterative
optimization-based attack to Mozilla's implementation DeepSpeech end-to-end,
and show it has a 100% success rate. The feasibility of this attack introduce a
new domain to study adversarial examples.
%0 Generic
%1 carlini2018audio
%A Carlini, Nicholas
%A Wagner, David
%D 2018
%K adversarial
%T Audio Adversarial Examples: Targeted Attacks on Speech-to-Text
%U http://arxiv.org/abs/1801.01944
%X We construct targeted audio adversarial examples on automatic speech
recognition. Given any audio waveform, we can produce another that is over
99.9% similar, but transcribes as any phrase we choose (recognizing up to 50
characters per second of audio). We apply our white-box iterative
optimization-based attack to Mozilla's implementation DeepSpeech end-to-end,
and show it has a 100% success rate. The feasibility of this attack introduce a
new domain to study adversarial examples.
@misc{carlini2018audio,
abstract = {We construct targeted audio adversarial examples on automatic speech
recognition. Given any audio waveform, we can produce another that is over
99.9% similar, but transcribes as any phrase we choose (recognizing up to 50
characters per second of audio). We apply our white-box iterative
optimization-based attack to Mozilla's implementation DeepSpeech end-to-end,
and show it has a 100% success rate. The feasibility of this attack introduce a
new domain to study adversarial examples.},
added-at = {2019-07-09T15:50:43.000+0200},
author = {Carlini, Nicholas and Wagner, David},
biburl = {https://www.bibsonomy.org/bibtex/24e54966c5e9ab676e59cd388decdf4c6/topel},
interhash = {644a13e2ca48714469b4bf9487daecfe},
intrahash = {4e54966c5e9ab676e59cd388decdf4c6},
keywords = {adversarial},
note = {cite arxiv:1801.01944},
timestamp = {2019-07-09T15:50:43.000+0200},
title = {Audio Adversarial Examples: Targeted Attacks on Speech-to-Text},
url = {http://arxiv.org/abs/1801.01944},
year = 2018
}