Neural Voice Camouflage technology can help to prevent eavesdropping by producing custom background noise, which thwarts artificial intelligence (AI) that captures and transcribes recorded voices.
The solution uses machine learning to alter audio so the AI, but not people, misinterpret sounds.
Columbia University researchers trained a neural network on hours of recorded speech so it can constantly process two-second clips of audio, masking what it predicts will be spoken next.
The researchers overlaid their system's output onto recorded speech as it fed into an automatic speech recognition (ASR) system; their technology boosted the ASR software's word error rate from 11.3% to 80.2%.
Said Columbia University’s Mia Chiquier, “Artificial intelligence collects data about our voice, our faces, and our actions. We need a new generation of technology that respects our privacy.”
View Full Article
Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA
No entries found