Audio Adversarial Examples: Targeted Attacks on Speech-to-Text (via twitter.com).

Given any audio waveform, we can produce another that is over 99.9% similar, but transcribes as any phrase we choose (…).