Targeted Audio Adversarial Examples
This is phenomenal:
We have constructed targeted audio adversarial examples on speech-to-text transcription neural networks: given an arbitrary waveform, we can make a small perturbation that when added to the original waveform causes it to transcribe as any phrase we choose. In prior work, we constructed hidden voice commands, audio that sounded like noise but transcribed to any phrases chosen by an adversary. With our new attack, we are able to improve this and make an arbitrary waveform transcribe as any target phrase.
The audio examples on this page are impressive — a little bit of background noise, such as you might hear on a telephone call with high compression, hard to perceive if you aren’t listening out for it. Paper here: https://arxiv.org/abs/1801.01944 (Via Parker Higgins, https://twitter.com/xor )(tags: papers audio adversarial-classification neural-networks speech-to-text speech recognition voice attacks exploits via:xor)